Temporal dynamics of mental imagery, craving and consumption of craved foods: an experience sampling study

Abstract

Objective: According to the elaborated intrusion theory of desire, an initial thought about a wanted substance is elaborated with mental imagery, which increases craving and the probability of consuming the substance. We used an app-assisted experience sampling approach to test this theory in the context of food craving and eating.Design: Overall, 221 females (mean age = 21 years; mean body mass index = 22) reported craving, mental imagery, and food consumption six times per day (2 h intervals) for seven consecutive days. Additionally, two traits (general food craving and imagery ability) were assessed.Main outcome measures: craving intensity, food consumption.Results: The probability of eating a craved food increased if the vividness of the mental food image and craving intensity increased two hours before – independent of trait food craving and trait imagery ability. We also found evidence of controlled eating behavior, with participants consuming the food they craved in only 38% of the cases.Conclusion: Mental imagery vividness and craving intensity predict consumption of craved food. The association between craving and eating might be stronger in individuals who struggle with controlling their eating behavior. Therefore, future studies should examine these relationships in overweight/obese samples or patients with eating disorders.

Keywords:

• Craving
• food consumption
• mental imagery
• experience sampling methodology
• multilevel modeling

1. Introduction

Food cravings are defined as an intense desire to eat a specific food (Pelchat, 2002). They are common experiences (Lafay et al., 2001) that can become maladaptive in some cases. For example, research shows that food cravings are associated with binge eating (Moreno et al., 2009), weight gain, and obesity (Verzijl et al., 2018). With overweight and obesity prevalence reaching enormous proportions (WHO, 2016), the need for a clear understanding of the processes that facilitate food cravings and food consumption is paramount to facilitate the development of targeted interventions.

Food cravings are not only associated with hunger (Hill et al., 1991) but also with a wide range of affective, cognitive, and situational triggers. For example, they are associated with negative mood (Hill & Heaton-Brown, 1994), expectations (Yeomans et al., 2008), and specific times of day (Reichenberger et al., 2018). Moreover, food cravings can also occur as a consequence of cognitive elaborations of initially intrusive thoughts about food (Kavanagh et al., 2005).

Research on cognitive processes that underpin cravings points to a key role of mental imagery in cravings. Mental imagery can be defined as a simulation or a recreation of a perceptual experience (Kosslyn et al., 2001). A cognitive model of cravings – the elaborated intrusion theory of desire (Kavanagh et al., 2005), places mental imagery in the heart of the craving experience. According to this model, cravings result from the elaboration of intrusive thoughts about the desired object through mental imagery (May et al., 2004, 2015). More intense and vivid imagery is associated with more intense craving, which in turn promotes food consumption.

Experimental and questionnaire studies that examined the role of imagery in food cravings support these assumptions and show that individuals create vivid mental images of craved food (May et al., 2008). More specifically, studies show that craving intensity is positively related to imagery vividness (Harvey et al., 2005; Kemps & Tiggemann, 2007; Shahriari et al., 2019). Additionally, studies that asked participants to recall the most recent craving experience showed that over 50% of respondents reported craving-related imagery, which significantly predicted craving intensity but not food consumption (Schumacher et al., 2019). Of note – studies mostly focus on specific craving-related imagery and consumption of specific foods. Nevertheless, psychometric studies find that specific cravings correlate highly with general cravings and can be understood as dimensions of a higher order construct of general food cravings (White et al., 2002).

The studies mentioned above offer important insights into the relationship between imagery processes, food cravings, and eating. Studies that assess food cravings and consumption in naturalistic environments are critical because they are less affected by heightened awareness of observation, which influences intake behavior in the laboratory (Robinson et al., 2015). Most of the studies in naturalistic settings, however, assess food craving (and subsequent food consumption) with a cross-sectional design by asking participants to recall a recent craving episode and the consumption that followed (e.g., Schumacher et al., 2019; Tiggemann & Kemps, 2005) or by using paper-pencil measurements (e.g., Hill & Heaton-Brown, 1994). While informative, such methods are limited by drawbacks, such as low compliance rates and under-reporting of cravings (Berkman et al., 2014). Additionally, assessments of craving (and food consumption) at one specific time point do not provide any insight into the temporal dynamics of psychological processes.

The current study addresses these limitations by using experience sampling methodology (ESM; also referred to as ecological momentary assessment; Kirtley et al., 2021). The participants completed a series of brief questionnaires several times a day to capture in-the-moment reports. Besides the advantages presented beforehand, ESM also reduces recall bias and increases ecological validity (Myin-Germeys et al., 2018). Some examples of constructs studied concerning food cravings using a similar methodology have been snack-related thoughts (Richard et al., 2017) and experience of stress (Reichenberger et al., 2021). However – to the best of the authors’ knowledge, the role of imagery has not yet been examined with intensive longitudinal approaches.

Food cravings in everyday life are of specific relevance mainly because they are associated with increased food consumption (of mostly calorically dense foods), thereby contributing to weight gain and obesity (Boswell & Kober, 2016). Literature on the relationship between craving and food intake is substantive. Experimental studies usually expose participants to food cues and monitor craving and intake following food cue exposure. These studies consistently show that exposure to food cues increases cravings, which in turn lead to increased food intake. For example, a recent study by Sinha et al. (2019) found a correlation of .82 between general cravings for high-palatable foods and the subsequent caloric intake. Additionally, a study by Chao et al. (2014) examined food cravings and intake of the respective type of food (i.e., high fats, sweets, starches, and fast food) and found moderate positive correlations (ranging from .25 to .51) for all food types. General food cravings also increase the risk of binge-eating episodes (eating an atypically large amount of food in a short period of time while feeling loss of control; Joyner et al., 2015). Thus, general cravings are linked to both the type of eating and the amount of food intake. For a meta-analytic review and a list of studies examining the impact of craving on eating behavior see Boswell and Kober (2016).

Moreover, previous studies found positive associations between state and trait food craving and consumption of the craved food (Meule & Hormes, 2015). It should be noted, however, that several situational and individual factors influence consumption of a craved food and that food cravings do not always result in subsequent consumption (Hill, 2007). For example, a study by Richard et al. (2017) found that trait-level variables, such as trait food craving, are important to consider when modeling food consumption. More specifically, they found that the association between craving intensity and consumption of snacks was stronger in high trait food cravers relative to low trait food cravers. In the context of mental imagery, trait imagery ability could also be an important variable to consider when modeling cravings and craving-related food consumption. Studies show that individuals differ significantly in the vividness, precision, and ability to generate mental imagery (Reeder, 2017). Moreover, previous experimental work has shown that state food imagery is positively related to trait imagery ability (Tiggemann & Kemps, 2005).

The present study aimed to expand the understanding of food cravings in everyday life by examining the role of imagery processes in food cravings and craving-related consumption, utilizing the advantages of experience sampling in capturing temporal dynamics. We first looked at what types of food participants craved. In line with previous work (e.g., Reichenberger et al., 2018) we expected that participants will mostly crave foods high in calories, such as sweets, salty snacks, and starchy foods. Based on previous experimental and cross-sectional work and the elaborated intrusion theory of desire (Kavanagh et al., 2005) we predicted that (1) measurement occasions with more vivid imagery will be accompanied with more intense craving. To further examine the temporal relationship between craving-related imagery, craving and food consumption, we predicted that (2) craving intensity and craving-related imagery vividness at a previous timepoint will positively predict craving-related food consumption. Lastly, we examined the role of trait food craving and trait imagery ability as possible person-level moderators of the relationship between imagery, craving and consumption. We predicted that (3) the relationship between craving-related imagery and craving intensity will be stronger for people with high trait food craving and high trait imagery ability. Additionally, we predicted that (4) for people with high trait food craving and high trait imagery ability higher craving and imagery at the previous timepoint will be associated with higher likelihood of food consumption.

2. Method

The data used in this study were drawn from a larger study on ‘Food cravings in everyday life’ that received approval from the institutional ethical committee (approval number 038-30-86/2020/4/FFUM). We only discuss the measures analyzed for the current study. Data for this study are available on the Open Science Framework (https://osf.io/rej2t/).

2.1. Participants

A total of 223 healthy women (i.e., with no history or current psychiatric diagnoses) were recruited for the study. We had the goal of recruiting at least 200 participants, which would result in 80% power to detect small effects (r = .15, α = .05), allowing for up to 25% attrition. We recruited through students’ mailing lists. Participants received course credit for completing at least 80% of ESM prompts. Two participants had poor compliance (i.e., < 50% of all ESM prompts completed) and were thus omitted from the final sample due to concerns about low-quality responding. The analyses were performed on a sample of 221 participants, aged between 18 and 32 years (M = 21.05; SD = 2.32) and an average self-reported body mass index of 21.94 kg/m² (SD = 3.06).

2.2. Questionnaires

Food cravings questionnaire – trait (FCQT-r; Meule et al., 2014). The questionnaire assesses the frequency and intensity of food craving experiences in general. Participants are asked to rate 15 items (i.e., ‘If I eat what I am craving, I often lose control and eat too much’) on a scale from 1 (never/not applicable) to 6 (always). Higher scores indicate that participants experience more frequent and intense food cravings. Internal consistency in the current study was high (Chronbah’s α = .94).

The Plymouth Sensory Imagery Questionnaire (PSI-Q; Andrade et al., 2014). The questionnaire assesses mental imagery vividness ability in several modalities. For the purposes of the current study, we used items assessing three study-relevant modalities: visual (e.g., ‘imagine the appearance of a cat climbing a tree’), gustatory (e.g., ‘imagine the taste of a lemon’), and olfactory (e.g., ‘imagine the smell of a rose’). The questionnaire, therefore, consisted of 15 items (5 items per modality), which participants rate on a scale from 1 (no imagery) to 7 (imagery as vivid as real-life). A higher score represents greater imagery vividness ability for each modality. All items can also be summed up for a total score. The internal consistency of the questionnaire was satisfactory for all modalities (visual: α = .81, gustatory: α = .78, olfactory: α = .85) as well as for the total score, Cronbach’s α = .89.

2.3. ESM measures

Participants completed all ESM questions every two hours for seven consecutive days (six times a day).

Current craving. Participants were asked to rate the intensity of their current craving (i.e., ‘How strong is your desire for specific foods in the current moment’) on a scale from 0 (no craving) to 100 (extremely intense craving). Participants were instructed that they should report their craving as specific if they crave either a specific taste of food (e.g., something sweet) or a specific type of food (e.g., chocolate). If they rated their current craving higher than 0, they were further asked to specify what type of food they were craving the most by choosing one of the following pre-specified food type categories: salty snacks (e.g., chips, pretzels), sweets (e.g., chocolate, cookies), fatty foods (e.g., burger, pizza, fries), starchy foods (e.g., bread, pasta, rice), dairy products (e.g., yogurt), vegetables/salad (e.g., tomatoes, carrots) and fruit (e.g., apples, berries). The food categories were derived from the Yale Food Addiction Scale 2.0 (Gearhardt et al., 2016) and extended by healthy food categories in line with previous research (Reichenberger et al., 2018). At this point, they also could indicate that they do not have a craving for a specific food. Since craving is defined as an intense urge to eat a specific food (e.g., Pelchat, 2002), only those participants who reported specific cravings provided additional information about the craving episode.

Craving-related imagery. If participants indicated a specific craving, they were asked to provide information about their current imagery related to that specific craving. More specifically, they rated the vividness of visual imagery (‘do you have a visual image of the food you crave in your mind?’), gustatory imagery (‘do you have a representation of the taste of the food you crave in your mind?’) and olfactory imagery (‘do you have a representation of the smell of the food you crave in your mind?’). Participants rated all imagery items on a scale from 0 (no imagery) to 100 (completely vivid imagery).

Consumption. At each time point, participants indicated whether they ate something since the last ESM prompt by choosing YES/NO. If they reported food consumption, they indicated the type of food they consumed by selecting all that applied from the list (multiple answers could be selected). The list included the same categories as for the craving, with the addition of ‘other’.

2.4. Procedure and ESM protocol

One day before starting with the study, participants took part in an online session (∼ 45 mins) during which they were trained on the ESM protocol, and the SEMA3 smartphone app (Koval et al., 2019) used to collect the data. After the online session, participants completed a set of online questionnaires (see 2.2. Questionnaires) and responded to two training prompts to check for possible problems (data were discarded). The ESM protocol started the following day, with participants completing seven consecutive days of experience sampling. The smartphone app signaled six times a day following an interval-contingent (i.e., fixed) design every two hours starting at 10 a.m. (10 a.m., 12 p.m., 2 p.m., 4 p.m., 6 p.m., and 8 p.m.). Participants could postpone the completion of the survey for up to 30 minutes. The survey elapsed after 30 minutes. Each participant, therefore, received a total of 42 signals over the week of participating in the study. At each signal, participants answered questions regarding their current craving, craving-related imagery (all pertaining to how they felt at the moment), and also reported food consumption (and information about the consumed food) since the previous signal. If participants did not experience craving, they answered neutral items that were matched with items related to food craving in terms of type and length of question. The questions were related to their current social activity (i.e., are you alone or in company), social network activity (i.e., did you spend any time on social networks since the last survey) etc.

2.4. Statistical analysis

Analyses were conducted using R (version 4.0.2; R Core Team, 2020). To account for the nested structure of our data, we conducted multilevel modeling using lme4 (Bates et al., 2015). To increase interpretability and to reduce convergence issues, we standardized all variables following previous work (e.g., Kalokerinos et al., 2019). When participants reported no specific craving (i.e., rated current craving intensity with 0 or did not specify a specific food they craved), the respective measurement occasions were disregarded because no corresponding data for craving-related imagery were available. To address possible reactivity (i.e., a change in craving due to participating in the study), the effect of measurement occasion (i.e., day of participation in the study ranging from 0 [day 1] to 6 [day 7]) on craving was tested at Level 1.

Craving and mental imagery were assessed momentarily, and food consumption (with the additional information about the food if consumption took place) was assessed retrospectively (in the interval since the last prompt). Therefore, when modeling food consumption, lagged predictors from the previous time-point (i.e., t-1, which corresponds to 2 hours before) were entered in the model, allowing us to test directionality and to measure more direct effects. Only ratings one signal before food consumption were considered. We excluded overnight lags.

We modeled craving-related imagery as a Level 1 predictor of craving intensity. At Level 2, we modeled trait food craving and trait imagery ability as predictors of the intercept (i.e., between individual differences in craving intensity) as well as moderators of the Level 1 slopes. Slopes were allowed to vary randomly. For food consumption, we first modeled craving-related imagery and craving intensity (both lagged by 1 measurement occasion) as predictors of food consumption. Because craving-related food consumption is the focus of the current study, we analyzed occasions when participants reported consuming food that matched the previously reported craving at t-1 (coded as 0 for no consumption and 1 for consumption). Because the outcome variable of consumption is binary, analyses were conducted using a mixed effects logistic regression analysis using the function glmer (generalized linear mixed model) of the lme4 package. Craving-related imagery and craving intensity (both lagged by 1 measurement occasion) were entered as Level 1 predictors. At Level 2 we again modeled trait food craving and trait imagery ability as predictors of the intercept and moderators of the Level 1 slopes. The inclusion of random slopes led to computational issues (i.e., models did not converge), therefore, random slopes were removed when modeling food consumption.

All Level 1 predictors were person-mean centered (i.e., we subtracted the mean of the person from each score), which allowed us to model within-person change. Level 2 predictors were grand-mean centered (i.e., we subtracted the grand mean from each score), which allowed us to model between-person effects.

3. Results

3.1. Compliance and reactivity

Out of the 9282 possible ESM prompts, participants responded to a total of 8671 prompts, resulting in a compliance rate of 93% (SD = 7%; range 52% − 100%). Since ESM can cause reactivity, we tested whether participants’ craving reports changed over the days of participation in the study. No effect of day was found on reported craving (β₁₀ = −0.005, p = .27), indicating that participating in the study did not alter craving reporting.

3.2. Overall frequency of cravings and craving-related food consumption

Participants reported cravings (desire to eat a specific food) on 3804 (44%) occasions. Out of all measurement occasions where craving was reported, participants reported most frequently craving sweets (n = 801 or 21% of all measurement occasions), closely followed by starchy foods (n = 737; 19%). This was followed by vegetables (n = 625; 16%), salty snacks (n = 606; 16%), fatty foods (n = 399; 10%), dairy products (n = 324; 9%) and fruit (n = 312; 8%).

Participants reported food consumption on 5252 occasions (61%). On the participant level, they reported food consumption between 8 and 37 times (M = 23.76, SD = 5.19) during the week. When participants reported craving at t-1, they reported consuming food on 2129 occasions (41%) at the following time-point (excluding overnight lags). The consumption matched their craving (e.g., they reported craving sweets and then consumed sweets after that) 1128 times (53% of all occasions when they reported consuming food following craving at t-1). If we also take occasions when they did not consume food following craving at t-1, the consumption matched their craving on 38% of all occasions. More information on food consumption in relation to craving reports at t-1 (also in terms of specific food categories) can be found in the Supplementary material.

3.3. Correlations between averaged craving intensity, imagery vividness, and consumption

We first examined correlations between craving intensity and craving-related imagery by averaging the variables across all measurement occasions (see Table 1).

Table 1. Correlations between averaged ESM measures and trait measures.

	1.	2.	3.	4.	5.	6.
Craving intensity	–
Visual imagery	.21**	–
Gustatory imagery	.18**	.87**	–
Olfactory imagery	.19**	.69**	.78**	–
Trait food craving	.12	.02	- .01	.01	–
Trait imagery ability	.01	.17*	.27*	.37*	- .06	–

Notes. ** p < .01, * p < .05.

All imagery modalities were positively correlated with craving intensity. Trait food craving was marginally (p = .07) associated with craving intensity. Trait imagery ability was significantly correlated with all modalities of imagery. Due to a high correlation between the imagery modalities (i.e., correlation coefficients above .69), we combined the different imagery modalities into one Level 1 variable (i.e., craving-related imagery vividness, which is the momentary average of all three imagery vividness modalities) and included this variable into further models examining the interplay between imagery vividness and other Level 2 variables for predicting craving and food consumption.

3.4. ESM analyses

The descriptives of all Level 1 and Level 2 variables included in the analyses are presented in Table 2. Descriptives for Level 1 variables were averaged across all measurement occasions when craving for a specific food was reported. Craving mean when also including measurement occasions when no specific craving was reported was M = 26.21 (SD = 32.76).

Table 2. Descriptive statistics of continuous Level 1 and Level 2 variables.

Variable	M	SD
Level 1 (measurement occasions)
Craving intensity (0 − 100)	54.96	29.30
Visual imagery vividness (0 − 100)	55.54	33.74
Gustatory imagery vividness (0 − 100)	53.02	33.49
Olfactory imagery vividness (0 − 100)	39.67	34.30
Imagery vividness (averaged) (0 − 100)	49.42	31.05
Level 2 (participants)
Trait food craving	2.59	0.89
Trait imagery ability	5.37	0.96

3.4.1. Imagery and craving

We examined predictors of craving intensity variability within participants based on craving-related imagery and included a momentary average of the three craving-related imagery vividness modalities as a Level 1 predictor. Afterward, we included two Level-2 predictors (i.e., between-participants; trait imagery and trait craving) to examine whether these person-level characteristics moderate the relationship between momentary craving-related imagery vividness and craving intensity. To examine simple relationships, each Level-2 predictor was modeled separately (see Table 3).

Table 3. Coefficients, standard errors (SE) and p-values of multilevel models with craving-related imagery as Level 1 predictor (model a) and trait food craving (model b) and trait imagery ability (model c) as Level 2 predictors of craving intensity.

Model	β (SE)	p
Only Level 1 predictors^a
Level 1
Intercept (β00)	.888 (.029)	< .001
Imagery (β10)	.348 (.019)	< .001
Trait food craving as Level 2 predictor^b
Level 1
Intercept (β00)	.888 (.028)	< .001
Imagery (β10)	.345 (.019)	< .001
Level2
Trait food craving (β01)	.067 (.029)	.021
Cross-level interaction
Trait food craving x imagery (β11)	- .026 (.019)	.169
Trait imagery ability as Level 2 predictor^c
Level 1
Intercept (β00)	.889 (.029)	< .001
Imagery (β10)	.352 (.019)	< .001
Level2
Trait imagery (β01)	.003 (.029)	.895
Cross-level interaction
Trait imagery x imagery (β11)	− .028 (.019)	.151

Notes. ^a Level 1 equation (equal for all models): craving intensity_ij = π_0j + π_1j (imagery) + e_ij.

^b Level 2 equation: π_0j = β₀₀ + β₀₁ (trait food craving) + u_0j; π_{1j =} β₁₀ + β₁₁ (trait food craving) + u_1j.

^c Level 2 equation: π_0j = β₀₀ + β₀₁ (trait imagery) + u_0j; π_{1j =} β₁₀ + β₁₁ (trait imagery) + u_1j.

When looking at the model with only Level 1 predictors (model a), the results showed that craving-related imagery co-occurred with craving intensity, with the positive coefficient showing that more vivid momentary imagery was related to a higher craving intensity – this was also the case when controlling for trait food craving and trait imagery ability (models b and c). Trait food craving was a significant predictor of the intercept (i.e., between individual differences in craving intensity), with individuals with higher trait craving reporting more intense craving. However, trait food craving was not a significant moderator of the relationship between imagery vividness and craving intensity. Trait imagery did not predict individual-level differences in craving intensity and did not moderate the relationship between craving-related momentary imagery vividness and craving intensity.

3.4.2. Imagery, craving, and consumption of craved food

We first tested the role of craving intensity and craving-related imagery vividness at the previous time-point as predictors of craving-related food consumption reported at the following time-point. The results are presented in Table 4 (see model a – only Level 1 predictors).

Table 4. Coefficients, standard errors (SE) and p-values of the multilevel models with lagged craving intensity and craving-related imagery as Level 1 predictors (model a) and trait food craving (model b) and trait imagery (model c) as Level 2 predictors of craving-related food consumption.

Model	β (SE)	p
Only Level 1 predictors^a
Intercept (β00)	− .356 (.087)	< .001
Cravingt-1 (β10)	.590 (.072)	< .001
Imageryt-1(β20)	.356 (.078)	< .001
Cravingt-1 x Imageryt-1(β30)	- .075 (.057)	.192
Trait food craving as Level 2 predictor^b
Level 1
Intercept (β00)	− .361 (.083)	< .001
Cravingt-1 (β10)	.566 (.059)	< .001
Imageryt-1(β20)	.344 (.075)	< .001
Cravingt-1 x Imageryt-1(β30)	− .070 (.055)	.201
Level 2
Trait food craving (β01)	− .004 (.087)	.962
Cross-level interactions
Trait food craving x cravingt-1 (β11)	− .008 (.058)	.891
Trait food craving x imageryt-1 (β21)	.144 (.083)	.083
Trait food craving x cravingt-1 x imageryt-1 (β31)	− .067 (.057)	.233
Trait imagery as Level 2 predictor^c
Level 1
Intercept (β00)	− .355 (.083)	< .001
Cravingt-1 (β10)	.564 (.060)	< .001
Imageryt-1(β20)	.341 (.076)	< .001
Cravingt-1 x Imageryt-1(β30)	- .071 (.056)	.201
Level 2
Trait imagery (β01)	.078 (.080)	.329
Cross-level interactions
Trait imagery x cravingt-1 (β11)	- .017 (.058)	.770
Trait imagery x imageryt-1 (β21)	.075 (.077)	.331
Trait imagery x cravingt-1 x imageryt-1 (β31)	- .027 (.057)	.636

Notes. Food consumption was coded as 0 (no consumption) and 1 (consumption).

^a Level 1 equation (equal for all models): food consumption_ij = π_0j + π_1j (craving_t-1) + π_2j (imagery_t-1) + π_3j (craving_t-1 x imagery_t-1) + e_ij.

^b Level 2 equation: π_0j = β₀₀ + β₀₁ (trait food craving) + u_0j; π_{1j =} β₁₀ + β₁₁ (trait food craving) + u_1j; π_{2j =} β₂₀ + β₂₁ (trait food craving) + u_2j; π_{3j =} β₃₀ + β₃₁ (trait food craving) + u_3j.

^c Level 2 equation: π_0j = β₀₀ + β₀₁ (trait imagery) + u_0j; π_{1j =} β₁₀ + β₁₁ (trait imagery) + u_1j; π_{2j =} β₂₀ + β₂₁ (trait imagery) + u_2j; π_{3j =} β₃₀ + β₃₁ (trait imagery) + u_3j.

As seen in Table 4, when including only Level 1 predictors, craving intensity and craving-related imagery vividness are both independently positively associated with food consumption at the following time-point, indicating that individuals were more likely to consume the craved food on occasions that were preceded by more intense craving and vivid imagery (see Figure 1). The interaction between momentary craving intensity and imagery vividness was not significant.

Figure 1. The relationship between vividness of the mental food image at t-1 and subsequent food consumption [a]; the relationship between craving intensity at t-1 and subsequent food consumption [b]. Consumption of craved food was coded 0 for no consumption and 1 for craving-matched consumption.

We then included two Level 2 variables as predictors of the intercept (i.e., individual differences in food consumption) and as moderators of the relationships between craving intensity, imagery vividness, and food consumption. The results of the two models are presented in Table 4 (see models b and c). Both momentary craving intensity and imagery vividness at the previous time-point remained significantly positively related to food consumption at the following time-point even after trait food craving and trait imagery ability were entered in the model. Trait food craving did not predict food consumption directly. It also did not moderate the relationship between craving-related imagery vividness and food consumption or craving intensity and food consumption. The same was true for trait imagery ability, which was not directly associated with food consumption, nor did it moderate the relationship between state imagery vividness, craving intensity, and food consumption.

4. Discussion

To the best of the authors’ knowledge, the current study is the first to assess the relationship between mental imagery, craving, and craving-matched consumption with an intensive longitudinal design. We used experience sampling methodology (ESM) to capture the temporal dynamics of the craving and consumption processes. Additionally, we examined trait food craving and trait imagery as possible moderators of the relationships.

Overall, imagery proved to be an important factor in both – craving intensity and craving-related consumption. Due to cravings being defined as an urge to eat a specific food (Pelchat, 2002), which can only be satisfied by consuming the food that is craved (Bruinsma & Taren, 1999), we focused on craving-related consumption (i.e., when the reported food matched the previously reported craving at t-1) relative to no consumption when examining the role of imagery and cravings in food consumption. This could also be viewed as the tendency for individuals to ‘give in’ to their craving vs. resisting the craving. We found that both craving intensity and craving-related imagery were significantly related to consumption of the craved foods. More specifically, when participants reported more intense craving and more vivid imagery at the previous time-point, they were more likely to report ‘giving in’ to their craving and consuming the food they were craving at the following time-point. Most commonly, this corresponded to consumption of starchy food, sweets, or salty snacks (see Supplementary materials for more detail) – in line with previous findings on snacking behavior (Richard et al., 2017). This also corresponds to the elaborated theory of desire (Kavanagh et al., 2005), which suggests that imagery promotes craving, which in turn promotes food consumption. Previous work was not able to corroborate these relationships. More specifically, a recent study (Schumacher et al., 2019) found that only craving, but not imagery, predicted craving-related consumption. Nevertheless, it should be noted that the study by Schumacher et al. (2019) used a cross-sectional design, asking participants to think of a recent craving episode, associated imagery, and consumption. As mentioned previously, such approaches do not permit separating the within- and between-person variance and are less suited for dynamic processes such as imagery, craving, and food consumption. Nonetheless, more research is warranted to substantiate our findings.

Neither trait food craving nor trait imagery ability proved to be directly associated with craving-related consumption. This is in line with studies that used more naturalistic approaches to studying craving and consumption (Richard et al., 2017), supporting the idea that food consumption is a dynamic construct driven by many factors (Hill, 2007). Additionally, trait food craving and trait imagery ability did not moderate the relationships between momentary craving intensity, imagery vividness and subsequent food consumption, indicating that the effect of craving intensity and imagery vividness is the same for people with high and low trait food craving and imagery ability.

Of note – we found evidence of controlled eating in our sample of (mostly) normal-weight women. More specifically, we found that participants varied substantially in the frequency of consumption (some reported only consuming food once a day). Additionally, we found that – when craving was reported at t-1 – participants consumed food at approximately 40% of the following time-points (with overnight lags excluded) and they consumed food that matched their craving on approximately 38% of all measurement occasions (see Supplementary material for more details). In normal-weight samples with no eating disorders, craving seems to have a limited impact on the actual consumption. A recent study (Roefs et al., 2019) compared craving-matched consumption study between normal-weight and overweight individuals and found that the two groups did not differ. However, the mentioned study assessed craving immediately before eating a meal and found that craving matched consumption on approximately 90% of all occasions. This is much higher than found in the current study and could be explained by the timing of the measurement – participants in the study by Roefs et al. (2019) reported about their craving (and consumption) right before they were about to eat and the food was in front of them. This could increase craving, since exposure to food leads to higher food cue reactivity (Jansen, 1998). Individuals struggling with controlling their food intake may be more responsive to spontaneously occurring cravings and ‘give in’ more frequently. For example, previous studies have found that successful inhibition of food craving is more effortful in obese compared to normal-weight individuals (Wang et al., 2020). Therefore, future studies should look at the relationship between specific cravings and the subsequent consumption of food that matches the cravings more closely and also include samples that are overweight/obese or display binge-eating.

In terms of the role of imagery in craving, we found that (when averaged across all measurement points) imagery vividness (including all sensory modalities) correlated positively with craving intensity, corroborating previous cross-sectional findings (e.g., Schumacher et al., 2019). On a momentary level, participants reported more intense craving when they experienced vivid imagery. This is in line with experimental studies showing that imagery vividness is related to craving intensity when participants are instructed to imagine a recent craving experience or their favorite food (Tiggemann & Kemps, 2005). Neither trait imagery ability nor trait food craving moderated the relationship between momentary imagery and craving intensity. Therefore, irrespective of trait variables, momentary craving-related imagery contributes to craving-related intensity, pointing to a crucial role of imagery in the craving experience – in line with the elaborated intrusion theory of craving (Kavanagh et al., 2005).

We did find that trait food craving was positively associated with craving intensity, indicating that high trait food cravers also experienced higher craving intensity. This is in line with previous experimental and questionnaire studies showing that trait craving is associated with craving intensity (Meule et al., 2014; Tiggemann & Kemps, 2005). Interestingly, trait food craving was only marginally associated with craving intensity when momentary craving intensity was averaged across all measurement occasions, highlighting the importance of capturing dynamics when modeling processes such as craving. On the other hand, trait imagery ability was not significantly associated with craving intensity. This is somewhat contrary to expectations since previous studies found that people with higher overall imagery ability also show stronger, more vivid food images at the moment (Schumacher et al., 2019), which would be expected to lead to more intense cravings overall. The positive association between state and trait imagery was also apparent in our data when looking at imagery vividness that was averaged across all measurement time-points. However, these associations were weak to moderate. Our data, therefore, indicate that momentary imagery is a more important predictor of momentary craving, relative to trait imagery ability.

The current results should be interpreted with the following limitations in mind. Firstly, we conducted the study on a sample of young women. This decision was made because young women experience more common food cravings (Hormes et al., 2014; Pelchat, 1997). Due to practical reasons, participants reported cravings and consumption by picking a food category. This prevented us from directly assessing caloric intake or the overall amount of food eaten, which would be valuable when interpreting results. Additionally, there is a possibility that participants did not report all the food that was consumed. Moreover, we did not collect information on current hunger. Previous studies found that hunger does not influence the relationship between craving, consumption and thoughts about food and was independent of craving experiences (Reichenberger et al., 2018). However, future studies could include hunger as a potential covariate.

Future studies could expand the findings by adopting an open-question response format, where participants specify the exact food they crave and also specify the type and amount of consumed food – allowing for more detailed assessment of food intake. Furthermore, future studies should focus on samples who struggle with controlling their intake, such as overweight/obese individuals or individuals with binge eating disorder. As mentioned before – we found evidence of controlled eating behavior in our sample of healthy women. Intake control could be motivated by different factors – it could indicate chronic dieting or a focus on maintaining a healthy lifestyle (Polivy et al., 2020) – something that should be taken into consideration in forthcoming research.

Taken together, our findings point to the important role of momentary imagery vividness in both craving intensity and subsequent food consumption. It seems to contribute to failures in resisting the craved food – in line with predictions of the elaborated intrusion theory of desire. These effects were not moderated by trait food consumption or trait imagery ability, indicating that momentary experiences of vivid imagery of the craved foods drive craving and also contribute to food consumption regardless of trait food craving and imagery abilities. Momentary craving intensity also predicted subsequent craving-related consumption (with a slightly higher effect size compared to imagery), irrespective of trait variables. Our findings support the elaborated intrusion theory of cravings and add upon previous research in important ways. We have shown that there is a direct link between imagery, craving and consumption of craved foods. Importantly, our results also highlight the importance of considering momentary dynamics of food cravings and consumption and the state factors contributing to both. In regard to practical implications of our findings – these results underscore the utility of brief imagery-based interventions (e.g., Schumacher et al., 2018) that can be used by participants on-demand at times where they experience strong cravings and/or vivid craving-related imagery.

Data availability

Data for this study are available on the Open Science Framework (https://osf.io/rej2t/).

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

The author(s) reported there is no funding associated with the work featured in this article.