Associations between everyday exposure to food marketing and hunger and food craving in adults: An ecological momentary assessment study

Food marketing in television and digital media negatively affects appetitive sensations and eating behaviour in children, but effects are less well understood for outdoor food advertising and adults. This research used Ecological Momentary Assessment (EMA) to explore associations between exposures to food advertising in various contexts (television, digital, outdoors) and adults ’ hunger and craving for highly advertised food categories. Over one week, participants provided ratings of cravings for types of food (fast food, soft drinks, snacks/ confectionery, other) and hunger on a smartphone app up to six times per day when they saw a food advertisement (reactive assessment) and at random intervals (random assessment). Fifty-four participants (70.4 % female; 21.24 ± 3.84 years) provided 1223 assessments (24.7 % reactive, 75.3 % random). Data were analysed in R using multilevel multivariable linear regression models. Participants reported feeling hungrier (X 2 (1) = 5.85, p = .016, Δ AIC = 3.9) and having stronger cravings (X 2 (1) = 20.64, p < .001, Δ AIC = 318.6) after seeing food advertisements vs. random assessments. This was driven by greater hunger following television advertising exposure vs. random assessments ( β = 1.58, SE = 0.61, p = .010, 95 %CIs 0.38 to 2.78), food advertising via digital devices or outdoors was not associated with hunger. Participants experienced stronger craving after seeing a food advertisement on television ( β = 0.52, SE = 0.19, p = .006, 95 %CIs 0.15 to 0.89), outdoors ( β = 0.39, SE = 0.12, p < .001, 95 % CIs 0.16 to 0.62) and in digital media ( β = 0.36, SE = 0.14, p = .012, 95 % CIs 0.08 to 0.64), vs. random assessments. Cravings were (largely) specific to the advertised food category. EMA can be effective for assessing food marketing associations in adults. The current study provides evidence that food marketing is associated with hunger and craving in adults, which may, with replication, have implications for public health policy.


Introduction
Obesogenic ('obesity-promoting') environments are driving poor diet quality and increased weight gain in child and adult populations across the world (Swinburn et al., 2011;Ng et al., 2014).These outcomes make a substantial contribution to the global burden of chronic disease, including type 2 diabetes, hypertension, various cancers, coronary heart disease and stroke (Benziger et al., 2016).It has been suggested that part of what makes the current food environment obesogenic is the increased presence of unhealthy food cues combined with eating behaviours and diets that diminish our ability to resist those cues (Martin & Davidson, 2014).
Food and non-alcoholic beverage (hereafter referred to as food) marketing is a notable contributor to proliferating environmental food cues.Food marketing has been found to be highly prevalent across multiple media and settings globally, and to be dominated by promotions for the least healthy products such as fast food, snacks, and confectionery (Boyland & McGale, 2022).In a recent review of over fifty studies assessing outdoor marketing, almost a quarter of advertisements identified were for foods, of which a majority (>60 %) were deemed unhealthy (Finlay et al., 2022).Exposure to food marketing has been demonstrated to have robust impacts on acute eating and related behaviours (such as preference and choice) in children (Boyland et al., 2022a).These effects are seen at a category level as well as through brand switching.That is, food marketing drives increased desire for products in a wider category (e.g., fast food in general), not just increasing selection of the specific brand/product advertised over alternatives (Cairns et al., 2013), and does not appear to be moderated by the medium/context of advertising (i.e., television, digital, packaging) (Boyland et al., 2022a).Similar effects on these outcomes in adults in experimental settings have not been consistently found (Boyland et al., 2016;Mills et al., 2013).However, there are cross-sectional associations between unhealthy food advertising exposure across multiple settings and diet-related outcomes (Buchanan et al., 2018) and odds of obesity (Yau et al., 2021) in adults.Broader sociocultural impacts, such as on dietary norms, shifts in food and drink category preferences, and the cultural values underpinning food behaviours have also been attributed to adults' food marketing exposure (Buchanan et al., 2018).
Evidence supports the notion that food marketing effects are cumulative across multiple exposures and comprise a multitude of outcomes including appetite-related sensations such as hunger, desire to eat, and craving (Kelly et al., 2015).While these sensations commonly co-occur and both are key determinants of food intake, they are considered conceptually different.Specifically, hunger has been defined as a homeostatic and biological state of acute energy deprivation or subjective anticipation of energy deprivation (or even more simply as the absence of fullness), whereas food craving is an intense desire or urge to consume specific foods (Reichenberger et al., 2018).In a recent survey 72 % of adult and adolescent respondents reported exposure to food marketing on popular digital platforms; 14 % reported craving the product after exposure, and 8 % made a purchase (Pollack et al., 2021).Experimental food advertising exposure can also drive greater desire to eat in adults, relative to control (non-food) advertisements, and therefore may be said to prime eating-related motivations (Boyland et al., 2017).Trait external eating (the tendency to eat in response to external food cues) has also been found to moderate the impact of food advertising exposure on adults' intake (van Strien et al., 2012) and food cue responsiveness predicts greater food marketing recall, craving, and purchasing behaviour in young adults (Pollack et al., 2022).These findings appear to suggest that food marketing's impact on food behaviours is consistent with food cue reactivity theory, whereby exposure to visual food cues such as promotional images triggers cue-induced craving (Boswell & Kober, 2016).There is robust evidence that craving systematically and prospectively predicts food-related outcomes including eating and weight gain in both children and adults (Boswell & Kober, 2016).
Most studies exploring associations with and effects of food marketing on eating and related outcomes are conducted in children rather than adults.Therefore, several gaps remain in our understanding of the impact of food marketing on adults, and the extant evidence base has some notable methodological limitations.Firstly, food marketing research with adults is primarily conducted in controlled laboratory environments, in contrast with the more naturalistic study settings used with children e.g., schools (Boyland et al., 2016).This may in part explain why such studies often do not identify significant effects of food advertising on adults' intake, as awareness that food intake is being monitored has been shown to be a demand characteristic in eating behaviour experiments (Robinson et al., 2014).Further, laboratory designs neglect the importance of examining outcomes in natural settings where the exposures typically occur (van Strien et al., 2012).Secondly, research into the impacts of outdoor forms of food marketing on eating-related outcomes is particularly sparse and, where present, usually relies on medium-long-term retrospective recall of exposure (Yau et al., 2021).In a recent scoping review (Finlay et al., 2022) only three studies were identified that explored associations between outdoor food marketing exposure and behavioural or health outcomes (whereas 53 studies measured levels of exposure).Two of those studies reported significant positive associations between outdoor food marketing exposure (either self-reported or inferred from monitoring activity) and adults' dietary intakes, and the third found no significant association between self-reported exposure and adult weight status (Finlay et al., 2022).Furthermore, to the authors' knowledge, no previous study has explored outdoor food marketing and appetitive sensations (Finlay et al., 2022).A greater understanding of dietary mechanisms that could explain the observed links between the obesogenic food environment and body weight, and a contextualised evidence base, is critical to support policy and practical interventions to curb the obesity epidemic (Kirk et al., 2010).Globally, many countries have implemented policies to restrict the advertising of unhealthy foods on television, but few restrict such marketing in digital media (Boyland et al., 2022b) or in outdoor settings (Chung et al., 2022).Differential exposures to unhealthy food marketing across socioeconomic groups may be contributing to health inequalities (Backholer et al., 2021), but where restrictive policies have been implemented, they have been found to be effective against key indicators such as household purchasing of unhealthy foods (Yau et al., 2022;Boyland et al., 2022b).
Ecological Momentary Assessment (EMA) is a real-time and ecologically valid measure of behaviour and drivers of behaviour in relevant environmental contexts, overcoming many of the limitations of labbased research (Masterton et al., 2023).This method has been used successfully to capture exposure to, and attitudes towards, tobacco advertising closer to the moment of exposure, rather than relying on retrospective recall (Rose et al., 2017).However, EMA designs are yet to be fully explored in the context of food marketing research, despite evidence that they can effectively detect within-person associations between environmental, contextual, and social factors, and food intake (Ghosh Roy et al., 2019;MacIntyre et al., 2021).
Therefore, the current study used EMA to explore the associations between everyday exposures to food advertising in various contexts (television, digital, outdoors) and young adults' hunger and craving for highly advertised food categories for a one-week period.

Participants
A formal power analysis was not conducted because of a lack of relevant information to inform such a calculation.Instead, we recruited as many participants as possible within the constraints of the available resources for reimbursement, ensuring that we recruited at least participants to ensure unbiased standard errors (Maas & Hox, 2005).Sixty-one (61) participants were originally recruited from the University of Liverpool and local community, via intranet and social-media advertisement and word of mouth.To be eligible, participants needed to be between 18 and 30 years old and fluent English speakers.They were not eligible if they had ever been diagnosed with any eating disorder or if they were currently receiving medication that could affect their weight or appetite.The study was approved by the University of Liverpool ethics committee (Ref.5488) and all participants provided informed consent prior to participation.Seven participants were excluded from the analysis (for reasons of exclusion see 'data cleaning'), leaving 54 participants in our analysis sample.

Dependent variables
Hunger and state craving were the primary outcomes of interest required to address the study aim.
Hunger was assessed using a visual analogue scale (VAS), with possible responses to the question "How hungry do you feel at the moment?" from 0 (not at all) to 100 (extremely).
State food craving: Participants answered the following questions using a visual analogue scale (VAS) from 0 (not at all) to (extremely); "How much do you crave [type of food] right now?".Types of food participants were asked for were fast food, soft-drinks, snacks/ E. Boyland et al. confectionery, or other.These measures were included to allow examination of both category-specific associations with food advertising (e.g., fast food advertising exposure being associated with craving for fast food) and a more generalised association of food marketing and appetitive sensations (e.g., food advertising being associated with increased craving for food categories beyond that featuring in the advertisement).

Control variables
These variables were not of interest for the study's aims but were controlled for as they could influence our primary outcomes.
Trait food craving was assessed with the short version of the Food Cravings Questionnaire -Trait (FCQ-T-reduced (Meule et al., 2014);), which includes 15 fixed-response statements (e.g.When I crave something, I know I won't be able to stop eating once I start; If I eat what I am craving, I often lose control and eat too much) and responses are given on a Likert scale from 1 (never) to 6 (always).Total possible scores range from 6 to 90, with higher scores indicating greater trait craving.FCQ-T-reduced demonstrated excellent internal reliability in our sample (α = 0.94).
External eating: Trait external eating (eating behaviour driven by external food-related cues, regardless of internal states such as hunger) was assessed with the External Eating Behaviour subscale of the Dutch Eating Behaviour Questionnaire (DEBQ (van Strien et al., 1986);).The sub-scale includes 10 fixed-response questions (e.g., If food tastes good to you, do you eat more than usual?If food smells and looks good, do you eat more than usual?) and responses are given on a Likert scale from 1 (seldom) to 5 (very often).Total possible scores range from 10 to 50, with higher scores indicating greater external eating behaviour.The sub-scale demonstrated acceptable internal reliability in our sample (α = 0.70).
Dieting status: Participants were asked to endorse one of the following statements about their dieting status; following a structured weight loss programme (e.g., Weight Watchers, Slimming World, etc.), dieting by yourself (e.g., watching your weight or trying to cut down), or not currently dieting or following any sort of weight loss programme.

Procedure
Participants visited a laboratory at the University of Liverpool where they were screened for eligibility and provided informed consent.They then completed the DEBQ, the FCQ-T-reduced, and the dieting status question on a standard desktop computer.The order of these questionnaires was counterbalanced across participants.At the end of the lab visit participants were provided with a smartphone to carry for one week (EMA phase) with a purpose-built application (app) installed.The app comprised the VASs assessing craving for specific types of food (fast food, soft-drinks, snacks/confectionery, and other) and hunger.
Participants were asked to access the app and complete the appetitive sensations VASs up to six times per day.
(i) Random assessments -RAs.The smartphones randomly generated three notifications per day between 09:00 and 21:00.Participants were asked to complete the assessments the moment they received the notification.If that was not possible, we asked them to do it as soon as possible thereafter.However, we did not impose any time-lag restrictions, nor did we provide participants with reminders about uncompleted assessments or ask if they had recently encountered advertising. 1(ii) Reactive assessments -ReAs.Participants were also asked to complete the assessments when they saw a food advertisement.
To make the number of assessments more manageable and increase completion rates, we asked them to respond to only one food advertisement in each of the following time frames: 09:00-13:00, 13:00-17:00, and 17:00-21:00.When participants opened the app to complete ReAs, they were also asked to report where they have seen the advertisement, choosing among TV (e. g. regular TV food advertisement or food product placement during a show), digital (e.g. a food advertisement popping up in social media), or outdoors (e.g. a food-advertisement poster on the bus station), as well as what type of food was advertised, choosing among fast food, soft-drinks, snacks/confectionery, or other.
After using the app for one week, participants were invited for a second visit to the University to return the phone, get debriefed and compensated (£20 shopping vouchers or course credits).Participants received full compensation for at least 85 % completion of the EMA phase (17-18 RAs out of possible 21), reduced compensation (50 % less) for at least 50 % completion (10-11 RAs) and no compensation for less than 50 % completion or if the phone was lost or damaged.ReAs were not considered when calculating completion rates as it was impossible to know whether participants may have seen an advertisement but neglected using the app.Of the 61 participants recruited, 49 (80.3 %) received full compensation, 10 (16.4 %) partial compensation, and 2 (3.3 %) no compensation.

Technical details
The smartphones used in the study were Alcatel Pixi 3, Samsung Galaxy Ace 3 GT-S7275R, or Doogee X10 and they were operating on Android systems.The smartphone app was developed on OpenSesame (Mathôt et al., 2012), an open-source python-based software to create experimental tasks and questionnaires.Computer-based questionnaires were completed on Qualtrics, an online platform for surveys.To generate RA notifications on the smartphones we used Randomly RemindMe, an app available for free on Google Play by James Morris Studios.

Data cleaning
Of the 61 participants, seven were excluded from the analysis for the following reasons; four participants completed less than 50 % of RAs, one participant reported leaving the smartphone at home and completing all assessments retrospectively when getting back, one participant was identified as older than 30 years of age after they completed the study, and one participant reported that in some cases they accidently pressed the button to complete a ReA although their intention was to complete a RA.
Prior to data analysis, we discarded any assessments that were initiated but not completed (e.g., opening the app but not completing any VAS), or that were completed outside the EMA schedule (e.g., RAs completed on the day participants were scheduled to return their phones, or ReAs completed after 21:00 2 ).If participants had completed more than three RAs on a day or more than one ReAs in the given time frames (by logging on to the app and selecting the wrong assessment type by accident), we deleted the excess number of assessments.A random number generator was used to select which assessments to delete (e.g. if two ReAs have been completed between 09:00 and 13:00, the earliest was assigned number 1 and the latest number 2 and one of them was randomly deleted).
Across participants, we identified occasions where a RA and a ReA appeared to have taken place at the same hour and minute of the day.Although it is possible that a random notification appeared exactly at the same time when participants saw a food advertisement, we believe chances are slim.A more reasonable interpretation might be that whichever assessment happened first acted as a reminder for the latter.For example, participants might have accessed the smartphone to 2 Although random notifications stopped at 21:00, RAs that were completed after 21:00 were retained as participants were asked to complete RAs as soon as possible after getting the notification.
E. Boyland et al. complete a ReA and noticed that a RA was still pending and completed it after the ReA.Alternatively, they might have got a notification to complete a RA, and this reminded them of an advertisement they saw earlier and retrospectively completed the ReA after the RA.However, any food-marketing exposure associations might have still been present if a RA was completed straight after a ReA, or the associations might have worn off if the ReA was completed sometime after the exposure.To resolve this, we examined the time of completion (including seconds) and deleted whichever assessment happened later.

Analysis
As the data had a hierarchical structure (sessions, clustered within days, within participants), we analysed the data with a multilevel model, which allows for dependent observations caused by clustering and unequal data points (Hayes, 2006, pp. 385-410, Quené andvan den Bergh, 2004).More specifically, we used multilevel multivariable linear regression models with fixed slopes and random intercepts.As we obtained sufficient data from more than 54 participants this ensured unbiased standard errors (Maas & Hox, 2005).All analyses were performed using the lme4 package in R.
Data showed a positive skew therefore we calculated the most appropriate approach to modelling the data using the gamlss package in R. All analyses suggested that a Box-Cox transformation was most appropriate to model the outcomes (barring overall craving although the Box-Cox transformation performed well so is reported here).Transformations were based on the exact lambda (derived using the MASS package in R) for each variable.
We used differences in the − 2 log likelihood along with differences in AIC values (ΔAIC) to examine the exact structure of our data (see Supplementary Materials, Table S1).Our outcome variables of interest were hunger, average craving across all four types of food (fast food, soft-drinks, snacks/confectionery and other) and food-specific cravings.For all outcome variables, a 3-level model (sessions within days within participants) fitted the data better than a single level model or any of the 2-level models (sessions within days or sessions within participants).
To examine the effect of food-marketing location and type of food advertised on craving and hunger, two categorical variables were derived from the existing data; context of exposure coded as no exposure, TV advertisement, digital advertisement, or outdoors advertisement, and content of exposure coded as no exposure, fast-food advertisement, soft-drinks advertisement, snack/confectionery advertisement, and other-food advertisement.

Hunger
Hunger was analysed with a 3-level model (see Table 1).Mean hunger per assessment (transformed) was β 0 = 9.45 (Standard Error -SE = 0.35).23.6 % of variance in hunger ratings was attributed to between participants differences (3rd level of analysis), 4.3 % was attributed to between days differences within the same participant (2nd level), while 72.1 % was attributed to between assessments differences within the same day (1st level).
To examine whether exposure to food marketing affected hunger, a 3-level regression model was fitted.First, we ran a control model with age, sex, external eating behaviour, trait craving and dieting status, which was a better fit than the null model (X 2 (6) = 16.12,p = .013,ΔAIC = 4.1).We then added type of assessment (RA or ReA) as a first level explanatory variable to this model.The model was significantly better than the control (X 2 (1) = 5.85, p = .016,ΔAIC = 3.9) and explained 28.7 % of variance in hunger ratings at the participant level, 4.46 % at the day level and 0.4 % at the assessment level.Hunger was greater at ReAs compared to RAs indicating that participants reported feeling more hungry after seeing a food advertisement compared to random assessments.The only other significant predictor in model was trait craving.
To examine whether the context of exposure (no exposure, TV, digital, or outdoors) affected hunger, a 3-level regression model was fitted using context of exposure as a first level explanatory variable.Content of exposure was inserted as dummy variables with no exposure serving as reference level.The model was compared to the same control model as above.The model was significantly better than the control model (X 2 (9) = 8.76, p = .033,ΔAIC = 2.8) explaining 28.8 % of variance in hunger rating at the participant level, 4.1 % at the day level, and 0.7 % at the assessment level.Hunger was greater at TV exposure compared to no exposure (β = 1.58,SE = 0.61, p = .010,95 % CIs 0.38 to 2.78).There were no differences between digital or outdoors exposure and no exposure (ps > .05).Results reveal that participants felt hungrier after seeing a food advertisement on TV compared to random assessments, while seeing a food advertisement on a digital device or outdoors had no association with hunger.The only other significant predictor was trait craving (β = 0.08, SE = 0.03, p = .021,95 % CIs 0.01 to 0.14).

Mean craving
Mean craving (transformed) across all types of food (fast food, softdrinks, snacks/confectionery, and other) was analysed with a 3-level model (see Analysis).Mean craving per assessment was β 0 = 4.85 (SE = 0.20).53.8 % of variance in craving ratings was attributed to between participants differences (3rd level of analysis), 5.5 % was attributed to between days differences within the same participant (2nd level), while 40.3 % was attributed to between assessments differences within the same day (1st level).
To examine whether exposure to food marketing affected craving, the same method was used as for the hunger analysis (see Table 2).The control model was a significantly better fit than the random intercept only model (X 2 (6) = 28.61,p < .001,ΔAIC = 16.6).Type of assessment (RA or ReA) was added as a first level explanatory variable to this model.This model was significantly better than the control (X 2 (1) = 19.27,p < .001,ΔAIC = 17.3) and explained 42.6 % of variance in craving at the participant level, 2.9 % at the day level and 1.6 % at the assessment level.Craving was greater at ReAs compared to RAs indicating that participants experienced stronger food craving after seeing a food advertisement compared to random assessments.Again, trait craving was the only other significant predictor in the model.

Food-specific craving (see Table 2)
Fast-food craving and soft-drinks craving were analysed with a 3level model (see Analysis).Mean (transformed) fast-food craving per assessment was β 0 = 4.46 (SE = 0.20).35.3 % of variance in fast-food craving was attributed to between participants differences (3rd level of analysis), 6.4 % was attributed to between days differences within the same participant (2nd level), while 58.3 % was attributed to between assessments differences within the same day (1st level).Mean (transformed) soft-drinks craving per assessment was β 0 = 4.99 (SE = 0.29).56.4 % of variance in soft-drinks craving was attributed to between participants differences (3rd level of analysis), 3.7 % was attributed to between days differences within the same participant (2nd level), while 39.9 % was attributed to between assessments differences within the same day (1st level).Mean (transformed) snacks/confectionery craving per assessment was β 0 = 5.47 (SE = 0.25).38.0 % of variance in snacks/ confectionery craving was attributed to between participants differences (3rd level of analysis), 4.8 % was attributed to between days differences within the same participant (2nd level), while 57.2 % was attributed to between assessments differences within the same day (1st level).
To examine whether the effect of food marketing on craving is food specific, we examined the association between content of exposure (no exposure, fast food, soft-drinks, snacks/confectionery, or other) and each type of food craving separately.
As with the previous analysis, we first fitted a 3-level model with covariates and then fitted a subsequent model using content of exposure as a first level explanatory variable.Content of exposure was inserted as dummy variables with no exposure serving as reference level.
Fast food: The control model for fast-food craving was better than the random intercept only model (X 2 (6) = 29.17,p < .001,ΔAIC = 17.2).Content of exposure was then added as a first level explanatory variable to this model.This model was significantly better than the control (X 2 (4) = 48.76,p < .001,ΔAIC = 41) and explained 44.5 % of variance in craving at the participant level, 9.1 % at the day level, and 3.8 % at the assessment level.
Fast-food craving was greater at fast-food and snacks/confectionery exposure compared to no exposure.Results reveal that participants experienced stronger fast-food craving after seeing a fast-food or snacks/ confectionery advertisement, compared to random assessments (with the former association being stronger), they also had greater fast food craving following soft drink and snack exposure although the magnitude of these associations were smaller than those for fast food exposure.External eating and trait craving were also associated with fast food craving.
Soft-drinks: The control model for soft-drinks craving was a better fit than the random intercept only model (X 2 (6) = 19.38,p = .004,ΔAIC = 7.4).Content of exposure was then added as a first level explanatory variable to this model.This model was significantly better than the control (X 2 (4) = 25.65,p < .001,ΔAIC = 17.6) and explained 31.8 % of variance in craving at the participant level, none at the day level, and 3.0 % at the assessment level.Participants experienced stronger softdrinks craving after seeing a soft-drinks advertisement compared to random assessments, as well as after seeing fast food commercials, although the association was larger for soft-drinks (snack/confectionary and other commercials showed no associations).Trait craving was also associated with soft-drinks craving.

Snacks and confectionery:
The control model for snacks/confectionery craving was a better fit than the random intercept only model (X 2 (6) = 32.25,p < .001,ΔAIC = 20.3).Content of exposure was then added as a first level explanatory variable to this model.This model was significantly better than the control model (X 2 (4) = 12.75, p = .013,ΔAIC = 4.7) and explained 47.9 % of variance in craving at the participant level, 14.8 % at the day level, and 0.3 % at the assessment level.Results reveal that participants experienced stronger snacks/confectionery craving after seeing a snacks/confectionery advertisement compared to random assessments, while seeing any of the other types of advertisement had no association with snacks/confectionery craving.External eating and trait craving also predicted snacks/confectionery craving.
Histograms showing the distribution of the main variables of interest (hunger, craving) are provided in the Supplementary Materials, along with graphical illustrations of all results and a further exploration of advertising effects with a random slopes analysis.

Discussion
The aim of the current study was to use EMA to explore the associations between everyday exposures to food advertising in various contexts (television, digital, outdoors) and adults' hunger and craving for highly advertised food categories.The main findings were that (i) participants experienced stronger hunger and cravings after seeing food advertisements compared to random assessments, (ii) associations were found for TV (hunger and craving), outdoor and digital (craving only) advertising exposure, and (iii) cravings were (largely) specific to the advertised food category.First, food advertisement exposure was significantly associated with adults' hunger and craving.This is consistent (albeit there are some caveats, discussed below) with the findings of a previous cross-sectional survey (Pollack et al., 2021) and laboratory-based experimental study (Boyland et al., 2017) of TV and digital media (respectively) food marketing and eating motivations in adults.The current findings also appear to support cue reactivity theory as an explanation of the mechanisms through which food marketing exposure exerts its effects (Boswell & Kober, 2016).To the authors' knowledge, this is the first study to have explored outdoor food marketing and appetitive sensations.Therefore, while there are no direct equivalents in the literature to compare current findings with, they are broadly consistent with the few studies that have shown significant and positive relationships between self-reported outdoor food marketing exposure and dietary behaviours in adults (Finlay et al., 2022).
Second, associations with both hunger and craving were found for TV food advertising, and associations with outdoor and digital food advertising exposure were seen for craving.This is broadly consistent with recent evidence from child populations which suggests that the magnitude of food marketing impact on eating and eating-related behaviours (such as choice and preference) is similar across different media (TV, digital, packaging) (Boyland et al., 2022a).A 2021 study also demonstrated associations between food marketing via leading digital platforms (Twitch and YouTube) and craving and purchase outcomes in US adolescents and adults (Pollack et al., 2021), and a prior review reported evidence of consistent relationships between digital marketing exposure and attitudes/behavioural intentions towards, and actual consumption of, food and other unhealthy commodities in adolescents and adults (Buchanan et al., 2018).However, the differential findings on craving (TV, outdoor, digital) and hunger (TV only) also warrant consideration.This may relate to participant perceptions of immediate food accessibility being different across the different settings; namely that a variety of foods are likely to be readily available when watching TV but only available after purchasing (which may or may not be immediately feasible depending on the specific location and availability of resources) when outdoors or out of the home, given that most time online is via portable digital devices (World Health Organization, 2016).Future studies should consider the role of food accessibility when comparing food marketing responsivity across media and settings.
Third, these associations are also consistent with indications that food marketing operates at a category level, not just in terms of brand switching (Cairns et al., 2013).This has implications for researchers in this field, namely that studies exploring the behavioural outcomes of food marketing exposure should ensure that both types of responses can be identified in a comprehensive design (i.e., that alternative branded items and broader category items should be available to choose/consume).It also has theoretical implications, as it provides support for social cognitive theory as a meaningful explanation for the pathway from food marketing exposure to consumer behaviour.This theory suggests that food marketing can influence behaviour directly through automatic processes, regardless of explicit brand beliefs or attitudes.A recent cross-sectional study found that attitudes towards unhealthy foods mediated the relationship between recall of digital food marketing and purchase/consumption outcomes in adolescents (Evans et al., 2023).Future research is needed to better elucidate the mechanistic pathways in both adults and children (Harris et al., 2009).
Previous evidence for the category-level effect is largely derived from studies with children (Halford et al., 2007(Halford et al., , 2008) ) so the current study adds to the body of evidence for this category level effect in adults as well the overall food marketing effect in this age group which is understudied relative to children (Boyland et al., 2016;Mills et al., 2013).Specifically, exposure to advertising for soft-drinks was associated with greater cravings for both soft-drinks and fast food (the former was stronger), fast food advertising exposure was associated with greater cravings for both fast food and snacks/confectionery (although again the former was stronger) and exposure to advertising for snacks/confectionery led to greater cravings for these products only.The influence of fast food advertising on cravings for snacks may reflect the trend towards a blurring of lines between the two product types (Lufkin, 2019).This is supported by a study that showed over a third of Australian adults eat fast food as a snack (Scully et al., 2008).Snack frequency is also positively associated with frequency of fast food intake in the transition from adolescence to adulthood (Larson et al., 2008).However, it is worth noting that the reverse effect was not found (advertising for snacks/confectionery did not affect craving for fast food).This finding may warrant further exploration.Further, soft-drinks are also a standard component of fast food meal bundles (Boyland et al., 2015) which may explain the association between soft-drinks advertising and fast food craving.
These findings may have implications for public health policy.Previously, direct effects on adults have been excluded from UK Government health impact assessments for proposed television and online food advertising policies due to a 'lack of conclusive evidence' for the relevant behavioural outcomes (UK Government, 2019).The current study provides an indication of associations with hunger and craving, rather than an effect size estimate for caloric intake (as is typically used in these assessments to project calories not consumed by children, and therefore weight not gained and the resulting health benefits (Tatlow-Golden & Parker, 2020;Brown et al., 2018)).However, it adds to an evidence base on adult responsiveness to food marketing that was described by the Government as 'far less developed than … for children' (UK Government, 2019), and provides a starting point for studies to build on by exploring behavioural outcomes.Additional evidence of effects of outdoor food advertising may also support Local Authorities who may be considering implementing restrictive policies (Chung et al., 2022;McKevitt et al., 2023) similar to those enacted by Transport for London (Meiksin et al., 2022) and others.
The study has also demonstrated the utility of EMA technology to address research questions pertinent to food marketing research and the design and implementation of policies to restrict food marketing across multiple media and settings.Notably, EMA methods lend themselves particularly well to studies of the impact of outdoor food advertising where there is a relative dearth of research (Finlay et al., 2022) and to exploring inequalities in exposure and diet-related outcomes (Backholer et al., 2021) where greater clarification is required.
A strength of this study is that it is, to our knowledge, the first to apply EMA methodology to explore the impact of food marketing exposure on diet-related outcomes.This approach facilitated an ecologically valid examination of the phenomenon in the natural settings in which food marketing is encountered and in real-time, with measurements of the outcomes far closer to the point of exposure than is would be achieved with other study designs.
The limitations of the study are that it is based on a relatively small sample of young adults in one location (Liverpool) so findings may not be generalisable to other settings or populations.More research is needed to determine whether the findings are transferable.Further, the method used in the current study only captured associations with consciously perceived food advertisements, which may have affected responding.As discussed at length in relevant theoretical models (Folkvord et al., 2016) and conceptual frameworks of health behaviours (Hollands et al., 2016), the level of conscious attending and cognitive processing applied to advertising may be important to responding.The advertisements used in reactive assessments were also self-selected, as participants are likely to have been exposed to many more than three food advertisements on the test days but were asked to only complete three ReAs to keep the participant burden manageable.They may have reported on the most salient (e.g., favoured brands) or memorable (e.g., famous brands; new or innovative content) advertisements which may have introduced bias in their data (although participant non-compliance or attrition due to excess burden would also have introduced bias).RA E. Boyland et al. measures were also not entirely random, as participants were allowed to complete them at a convenient time and not necessarily immediately upon receiving the prompt, again to minimise burden or intrusiveness.This may also have introduced bias (but as above, greater demands for immediacy may have resulted in more data loss).A further limitation (as noted in the methods) is that outcomes on RAs may have been affected by earlier advertisement exposure.We are not aware of any data to indicate what might be considered an appropriate time "window" for advertisement influence; therefore we could not account for this in our analyses.
For future studies, it may be beneficial to combine EMA approaches with methods that collect time stamped exposures such as mapping of food advertising in the street scape (Palmer et al., 2021) against participant location data, and/or screen capture of actual (not consciously perceived) digital food marketing exposure (Kelly et al., 2023) to further explore these phenomena.Incorporating a time-lag restriction may also be advantageous, as would increasing the number of random assessments (although this adds to the demands on participants and potentially also the study costs due to increased reimbursements for inconvenience).Studies using EMA to collect data on the impact of food marketing exposure on actual food behaviours in adults (particularly purchasing and intake) are also needed.
To inform future research, it may be relevant to consider differences in how food marketing is delivered between different media and settings, including those in the current study.Digital food marketing, unlike TV or outdoor promotions, tends to be highly embedded in entertainment content (WHO Regional Office for Who Regional Office for Europe, 2016).Cognitive processing of digital media marketing is hindered by this, as well as the tendency for media multi-tasking (Baumgartner et al., 2014).Theoretical models predict that this reduced cognitive processing would enhance, rather than reduce, the impacts (Folkvord et al., 2016) and there is some experimental evidence that the effects of food marketing are greater when participants are distracted (Zimmerman & Shimoga, 2014).However, digital marketing is also an extremely diverse advertising format, including (but not limited to) social media, video sharing and game-streaming platforms, and food delivery applications (Bragg et al., 2020;Horta et al., 2021;Pollack et al., 2020).Even within social media, food marketing can be experienced in several different ways including via content posted on accounts owned by food companies, sponsored posts from food brands appearing in a news 'feed', influencer marketing, or word-of-mouth marketing (also referred to as 'earned media') which often shares or mimics original brand content (Qutteina et al., 2019;Holmberg et al., 2016).While there is some evidence that the level of engagement (liking, sharing) with food marketing affects its behavioural impact in children (Baldwin et al., 2018), and that peer endorsement of marketing content may also be important (Alhabash et al., 2015), this research is in its infancy.EMA methodologies could usefully be applied in future research to the challenge of disentangling the impact of different forms of digital food marketing exposure in real-world settings on diet-related outcomes.

Conclusions
This study used an EMA approach to explore associations between everyday exposures to food advertising and adults' hunger and craving for highly advertised food categories.We found that participants experienced stronger hunger and cravings after seeing food advertisements compared to random assessments, associations were found for TV (hunger and craving) and outdoor and digital advertising exposure (craving only), and cravings were (largely) specific to the advertised food category.This may have public health policy implications, but further research is needed to replicate this effect and extend to other primary outcomes (e.g., dietary intake).

Table 1
Regression models predicting hunger and craving.

Table 2
Regression models predicting craving of specific food and drinks.