1

Assessing individual motivation for consuming diverse food groups

over a two-week period

Johanna Werner1

Juliane Kloidt1

Lawrence W. Barsalou1

1 School of Psychology and Neuroscience, University of Glasgow

30th of November 2022

Please address correspondence to: Johanna Werner, School of Psychology and

Neuroscience, University of Glasgow, 62 Hillhead Street, Glasgow G12 8QB, United
Kingdom; j.werner.1@research.gla.uk
 2

Abstract

When attempting to lose weight or adopt a more sustainable diet, most people

struggle considerably to permanently change their eating behaviour. To develop

novel interventions needed to help participants achieve their goals, it is first

necessary to understand what motivates individual consumption. In the present

study, we assessed individual consumption motivation in 70 participants for 16

diverse food groups across 5 time points over a two-week period. Participants were

recruited to take part in an online survey. Across time points, participants rated the

16 food groups 5 times, first, on 2 dependent variables (consumption frequency and

portion size), and second, on 6 potential motives that could predict consumption

(enjoyment of taste, fillingness, healthiness, automaticity, emotional satisfaction, and

self-identity). The two dependent variables were later combined by calculating their

product to create a single dependent variable for consumption. Of interest was firstly

to assess whether implicit learning effects emerged, specifically an increase in

correlation strength between dependent variable and predictors from timepoint 1 to

timepoint 5. Second, using G-Theory and intraclass correlations, we assessed the

stability of each individual’s consumption motives over time. We also established the

motives that best explained each individual’s consumption and how much variance

they explained. Lastly, we assessed how much insight individuals have into their

consumption motives. We found significant implicit learning effects in the association

of consumption with healthiness, self-identity and filligness. Across the five time

points, participants consumption motives remained remarkably stable. We observed

large individual differences in motivation for consumption across participants. Finally,

participants exhibited little awareness of their consumption motives as captured by

their implicit motivation judgments. To our knowledge this is the first study to

establish the stability of situated consumption motivation over time. Our results
 3

furthermore provide possible starting points for novel interventions, empathising the

importance of individualised instead of ‘one size fits all’ approaches.

 4

Food has a unique place in our life. The specific kinds of food we consume

have an impact not only on our physical health (Ignarro et al., 2007; Willett, 1994),

but also on our mental health (El Ansari et al., 2014) and even our cognition

(Spencer et al., 2017). Consistently overconsuming foods results in excessive body

fat, which is further associated with numerous negative health outcomes, including

various forms of cardiovascular disease and cancer (Calle & Kaaks, 2004).

Although the hormones ghrelin (Higgins et al., 2007) and leptin (Cousino Klein

et al., 2004) act as physical signals to motivate food consumption, eating is a much

more complex process than the pure satisfaction of physical signals. For example,

other motivators not related to hunger such as stress (Groesz et al., 2012), emotion

(Reichenberger et al., 2020), norms (Higgs, 2015), and identity (Bisogni et al., 2002)

can influence eating behaviour extensively.

Food production in general, and especially meat production, is a substantial

contributor to global greenhouse emissions (O’Mara, 2011), that play a major role in

global warming (Al-Ghussain, 2019). Consequently, food choice not only shapes an

individual’s long-term health and wellbeing but can furthermore affect planetary

health (Birt et al., 2017). To make their eating more sustainable, people will have to

decrease their meat consumption in favour of a more plant-based diet (Sabaté &

Soret, 2014). Despite a considerable increase in media attention, most people

remain largely unaware of the link between their individual food choices and climate

change, again especially related to their meat consumption (Macdiarmid et al., 2016;

Tobler et al., 2011).

Furthermore, when people try to shift towards more sustainable diets,

numerous obstacles, such as negative social feedback (Wehbe et al., 2022), must be

overcome. Novel interventions are needed to help people adopt more sustainable
 5

diets. One possible starting point for developing such interventions can be found in

establishing an individual’s motivations for eating. Vainio et al. (2016), for instance,

found that the importance of eating motives was different for people actively

changing their diet compared to those with established diets, implicating the

importance of eating motives in behaviour change.

In summary, novel eating interventions are needed help people achieve their

dietary goals. To work towards this aim, it is fundamental to first fully understand the

underlying motives, not only for meat consumption, but for consuming all types of

food. It is also important to assess how much eating motivation varies across

individuals, how stable it is over time, and how much insight people have into it.

The Situated Assessment Method (SAM2)

The Situated Assessment Method (SAM2) is a psychometric approach that

evaluates individual differences in a target behaviour along two dimensions of

situatedness (Dutriaux et al., 2021). First, SAM2 assesses the target behaviour in the

situations where it occurs, rather than abstracting across situations as in most

psychometric instruments. Second, it evaluates the target behaviour across all

phases of the Situated Action Cycle, including the environment, self-relevance, affect,

action, and outcomes (for a detailed description of the Situated Action Cycle see

(Barsalou, 2020; Dutriaux et al., 2021). To understand eating, for example, a SAM2

approach would assess food consumption in the situations where they occur and

would assess it across all phases of the Situated Action Cycle.

SAM2 has previously been used to study a range of diverse target behaviours

such as habits (Dutriaux et al., 2021), eating behaviour (Werner et al., 2022), and

stress (Pedersen et al., 2022). The Pedersen et al. (2022) study is of particular

interest here. In Pedersen et al., (2022) stress was assessed once a week at three
 6

timepoints over a two-week period. At each time point, participants received covid-

related situations (e.g., worrying about catching the virus, taking public transport) and

evaluated each one on two dependent variables (eustress and distress) and on

potentially relevant factors from the Situated Action Cycle known to influence stress

(e.g., expectation violation, threat, coping effectiveness). As expected, based on

previous SAM2 studies, participants showed large individual differences in the factors

that predicted their distress and eustress. Remarkably, however, and surprisingly,

both distress and eustress became increasingly correlated with the stress predictors

across timepoints. For example, judgments of both distress and eustress correlated

more highly with expectation violation at the third timepoint than at the first. Similar

increases in predictive strength occurred for the other predictors as well.

These results implicate implicit learning effects across timepoints as

participants evaluate stress and its predictors. In the process of evaluating stressful

situations for distress, eustress, and predictive factors from the Situated Action Cycle,

relations between these measures grow stronger. Building on those results, one aim

of the current study was to assess whether the same implicit learning effects occur

when people assess their food consumption over time. Do motivational factors that

predict consumption (e.g., healthiness, taste) correlate with consumption increasingly

as participants evaluate them across multiple timepoints?

Previous SAM2 work on consumption motives

The current study also built on previous SAM2 work that investigated motives

for food consumption, such as healthiness, fillingness, and sweetness (Werner et al.,

2022). For this previous project, we used SAM2 to establish individual differences in

consumption motives at a single timepoint. Our first aim was to assess how much

consumption motives varied across individuals, along with how much variance they
 7

explained in their consumption frequency and desire. We were also interested,

however, in how stable these motives were across different eating situations

(breakfast, lunch, dinner, snack).

In the last of three studies, 204 participants rated 177 foods, each situated in 1

of 4 eating situations. Specifically, each participant evaluated each food on two

dependent variables (consumption frequency and desire) and on ten consumption

motives sampled from the Situated Action Cycle (healthiness, fillingness, sweetness,

bitterness, affordability, automaticity, self-identity, social connectedness, emotional

satisfaction, and situational transport).

Large individual differences emerged in motivational profiles (i.e., in the

specific motives that predicted consumption frequency and desire) for each

individual). Nevertheless, the ten predictors tended to explain large amounts of

variance for each individual’s consumption frequency and desire (median explained

variance across individuals of 59% and 66%, respectively). Surprisingly an

individual’s motives tended to be highly stable across the four eating situations. In

other words, an individual’s motives in one eating situation tended to predict their

motives in another eating situation. Lastly, we assessed how much insight

participants showed into what predicted their behaviour and found little agreement

between these assessments and prediction profiles in their SAM2 data. Participant’s

beliefs about their eating motives diverged considerably from their situation- and

food-specific judgments of eating motives.

Design of the current study

A remaining question we had been unable to address in our previous study

was how stable eating motives are over time. To assess stability over time, it is

necessary to have participants repeatedly assess eating motives at multiple time

 8

points. Thus, a primary purpose of the current study was to assess eating motives

on multiple occasions, in this case, over five time points every few days across a two-

week period. Doing so further allowed us to assess the implicit learning effects

observed in the multi-timepoint stress study described earlier (Pedersen et al., 2022).

We couldn’t, however, simply repeat Werner et al.’s (2022) study design at

multiple time points, because its design was much too large. Specifically, it assessed

177 foods in 4 eating situations on 2 dependent variables and 10 predictors, with all

this data collection requiring 3 sessions of 45 minutes over the course of 2 weeks. In

the study here, we needed to distil this original assessment into a single session

taking participants only about 15 minutes to complete.

We therefore decided to try a new approach. Instead of asking individuals to

evaluate a large number of specific foods (n=177), we asked people to evaluate how

much food they consumed for each of 16 food groups that cover all foods

comprehensively (e.g., fruits, vegetables, breads, unprocessed meat). For each food

group, participants were asked to remember the foods they had consumed from the

group over the past few days, and then to estimate how frequently they’d consumed

these foods and the typical number of portion sizes consumed.

To further distil the study design, we decided not to include specific eating

situations. This was further supported by the observed lack of variation in an

individual’s eating motivation across situations in Werner et al. (2022). Based on

those results, measuring motivation in specific situations here would not have been a

good use of time and resources. So, we simply asked participants to evaluate how

much they had consumed each of the 16 food groups across eating situations over

the previous two days.

 9

Finally, we further distilled the study design by reducing the number of motives

assessed. Results from Werner et al. (2022) informed the motives included in the

current study. Because automaticity, emotional satisfaction, and self-identity strongly

predicted consumption frequency and desire in Werner et al., we included them here.

In contrast, sweetness and bitterness only correlated weakly with consumption

frequency and desire in Werner et al. (2022), although participants believed that they

were strongly associated with their consumption behaviour. To capture the taste

dimension in the current study, we therefore decided to simply include a measure of

how much participants had enjoyed the taste of each food group over the past two

days. Although healthiness and fillingness did not correlate with consumption

frequency and desire as highly as automaticity, emotional satisfaction, and self-

identity in Werner et al. (2022), they tended to be moderately important, and

participants again believed that they were highly associated with their consumption.

For these reasons, we decided to also include them in the present study. In

summary, we included six predictors of consumption frequency here: enjoyment of

taste, fillingness, healthiness, automaticity, emotional satisfaction, and self-identity.

To measure our dependent variable—consumption of a food group over the

past two days—we assessed two facets of consuming the food group: frequency and

typical portion size. Specifically, we first asked people to estimate how many times

they had consumed the food group over the past two days (consumption frequency).

Across these occasions, we then asked them to estimate the typical number of

portions consumed at each, where a portion was defined as the amount equivalent to

one hand-full (a widely used method in dietary recommendations; (Brown et al.,

2021)). To compute the consumption amount for the food group at this time point, we

simply took the product of an individual’s frequency and typical portion size

estimates. If, for example an individual had consumed sweets on three occasions
 10

over the past two days with a typical portion size of two handfuls, their consumption

of sweets at this time point was specified as six portions.

Overview and predictions

The aim of the present study was to assess individual differences in

consumption motives of diverse food groups, to measure the stability of consumption

motives over time, and to explore whether we could detect implicit learning effects.

The study built on and extended our previous findings into individual differences in

eating motivation (Werner et al., 2022), as well as implicit learning effects over time

for stress (Petersen et al., 2022).

At 5 separate time points during the two-week study duration, participants

were asked to evaluate 16 diverse food groups on 6 eating motives as well as the 2

facets of our dependent variable (consumption frequency and typical portion size),

whose product constituted our dependent variable of consumption. Three of the five

timepoints assessed consumption over two weekdays (Tuesday and Wednesday),

with the remaining two timepoints assessing consumption over the weekend

(Saturday and Sunday). The length of two days was chosen to ensure the entirety of

the weekend (Saturday and Sunday) was captured, whereas the two weekdays

(Tuesday and Wednesday) were chosen to be of an equivalent length to the

weekend but in the middle of the week. At each timepoint, we also asked

participants to self-assess how much they believed that each of the six predictors had

influenced their food consumption over the previous two days. Notably, at each time

point, participants were repeatedly instructed to only recall their experience of the

past two days when making their judgments.

Based on previous results, we expected to observe implicit learning effects,

namely, the association of consumption with the six predictors was predicted to
 11

increase from timepoint 1 to timepoint 5. We furthermore expected to observe large

individual differences in the motives associated with consumption. Furthermore, we

predicted the consumption motives that were important for each individual would

explain a large amount of variance in their consumption. Lastly, we expected little

agreement between an individual’s beliefs about their consumption motives and their

actual motives associated with their consumption in the SAM2 data.

Methods

Participants

A total of 79 UK residents (48 Female, age M = 37.94, SD = 12.7, BMI M =

26.91 SD = 6.42) were recruited via the Prolific platform (www.prolific.co).

Participants were eligible to take part if they were between the ages of 18 and 70,

spoke English fluently, and had previously taken part in at least 10 other Prolific

studies with an acceptance rate of 95%. Participants were informed that they should

only participate in the study if they could complete all five sessions (failing to

complete any session would lead to exclusion from the remaining study and without

monetary compensation for previous participation). Only the 70 participants who

completed all five sessions were included in the analyses to follow. Eight out of

original 71 participants failed to complete all five timepoints in the first collection

cycle, despite multiple attempts by the researcher to reach out via email during the

data collection timepoints, indicating that participants either simply forgot to respond

in time or decided not to continue with the study (2 dropouts in T2, 4 dropouts in T3,

1 dropout in T4, 1 dropout in T5). The missing participants then replaced with 8 new

participants during a second collection cycle during which one participant dropped

out after T3, resulting in complete data sets for 70 participants. After successfully
 12

completing all five sessions, participants were paid 15£ for their participation (for a

total of about 1.5 hours time across the 5 sessions).

Materials

Food groups. To fully capture participants’ eating experiences, whilst also keeping

the duration of the survey manageable, participants were asked to report their

consumption of food groups instead of specific foods. Informed by nutritional

websites and other relevant studies, we identified 16 distinct food groups that we

expected would cover most individual diets (Table 1). The food groups included both

meats and plant-based products to cover carnivore, vegetarian, and vegan diets.

When participants were asked to assess a food group on one of the dependent

variables or on any of the six predictors, they were always presented with the food

group’s name, along with a few examples of foods belonging to it (as Table 1

illustrates). For instance, the food group “Nuts / Seeds” was illustrated with “raw

almonds, walnuts, sunflower seeds, chia, etc.” Examples like these ensured that

participants understood what foods were included in each food group and aided

memory of foods consumed from it over the past two days.

Dependent variables. Instead of measuring the dependent variable, consumption,

directly by asking how much of a food group participants had consumed during the

past two days, we measured two of its facets separately (consumption frequency and

portion size). Doing so allowed us to preserve potentially interesting individual

differences (eating small portions frequently on multiple occasions versus eating one

large portion in a single sitting). Furthermore, we thought that having participants

remember all eating occasions first before judging the typical portion size might

facilitate accuracy.
 13

Table 1
Food groups evaluated by participants and corresponding examples used to illustrate them

Food Group Examples

Fruit apples, strawberries, bananas, oranges, dates etc.

Vegetables broccoli, tomatoes, lettuce, potatoes, carrots etc.
Beans Pulses Legumes kidney beans, lentil, chickpeas etc.
Nuts Seed raw almonds, walnuts, sunflower seeds, chia etc.
Unprocessed Meat unprocessed chicken, beef, pork turkey etc.
Processed Meat ham, sausages, bacon, salami etc.
Fish Seafood salmon, tuna, cod, mussels, squid, prawn etc.
Dairy Egg cheese, butter, yogurt, cream cheese, sour cream etc
Plant-based Meat tofu, tempeh, quorn, meat-free sausages, mince etc.
Plant-based Dairy Egg vegan cheese, egg, yogurt etc.
Bread toast, bread rolls, Yorkshire pudding, wraps etc.
Pasta Rice spaghetti, couscous, white rice, brown rice, rice noodles etc.
Cereal oatmeal, granola, cornflakes, etc.
Sweets chocolate, chocolate bar, toffee, ice cream, cake, pie, biscuits, cookies etc.
Salt Snack crisps, salted nuts, crackers etc.
Fried Food chips / fries, fried fish, fried chicken, onion rings etc.

Specifically, at each time point, participants first indicated on how many

occasions they had consumed a food group (a discrete scale ranging 0 to 10

occasions) over the past two days. Participants then estimated the typical portion

size consumed across these occasions in handfuls (a continuous scale ranging from

0 to 5 handfuls). For the typical portion size, it was explained to participants that if

they had consumed a food group on two occasions in total, and that on the first

occasion their portion size was two handfuls whereas on the second occasion their

portion size was four handfuls, they should report a typical portion size of three.

Finally, to create our dependent variable, consumption, we took the product of

consumption frequency and typical portion size. If, for example, a person had
 14

consumed sweets on 6 occasions during the past two days and reported a typical

portion size of 2 handfuls during the past 2 days, their consumption score for sweets

was scored as 12 handfuls.

Predictors. The six predictors included were informed by our previous study on

individual differences in eating motivation (Werner et al., 2022), and, as previously

discussed, chosen to cover a broad range of eating motivations. In total six

predictors were included: enjoyment of taste, fillingness, healthiness, automaticity,

emotional satisfaction, and self-identity (Table 2). Importantly, when assessing a

predictor, it was stressed to the participant that they should only think of their

experience the past two days (e.g., “how much did you enjoy the taste of fruits and

nuts the past two days?”). Table 2 presents the eight judgment questions (two

dependent variables and six predictors). Predictors were always presented in the

same order to all participants across the five timepoints, whereas the order of the 16

food groups within a predictor’s rating block were randomised. Participants

responded to all predictors on a continuous scale, ranging from 0 to 10 (see Table 2

for the corresponding scale labels).

Procedure

The study was part of a larger project investigating eating and covid stress,

running online over a two-week period. Over this period, participants performed five

sessions at five time points, each assessing the eating experience of the previous

two days. Timepoints one, three, and five took place on Thursdays and captured the

previous two weekdays (Tuesday and Wednesday). Timepoints two and four took

place on Monday and captured the weekend (Saturday and Sunday).

 15

Table 2
Dependent variables and predictors, including the judgment question, scale, labels, and
ICC2 for each measure.

Variable name Evaluative question ICC2

Frequency (DV) On how many occasions did you consume FOOD GROUP the past 2 days? .40
(Discrete 0 to 10) (0;1;2;3;4;5;6;7;8;9;10)
Portion Size (DV) Across those occasions what was the typical number of portions you .32
consumed? [ 1 portion = 1 cupped hand]
(Continuous 0 to 5) (0; 1; 2; 3; 4; 5)
Consumption (DV) Product of frequency and portion size .30

Enjoy Taste How enjoyable was the taste of FOOD GROUP for you the past 2 days? .23
(Continuous 0 to 10) (Not at all enjoyable; somewhat enjoyable;
moderately enjoyable; highly enjoyable; extremely enjoyable)
Fillingness How filling was consuming FOOD GROUP for you the past 2 days? .24
(Continuous 0 to 10) (Not at all filling; somewhat filling; moderately
filling; highly filling; extremely filling)
Automaticity How automatic was consuming FOOD GROUP for you the past 2 days? .24
(Continuous 0 to 10) (Not at all automatic; somewhat automatic;
moderately automatic; highly automatic; extremely automatic)
Healthy How healthy was consuming FOOD GROUP for you the past 2 days? .62
(Continuous 0 to 10) (Not at all healthy; somewhat healthy;
moderately healthy; highly healthy; extremely healthy)
Emotional Satisfaction How emotionally satisfying was consuming FOOD GROUP for you the past .22
2 days?
(Continuous 0 to 10) (Not at all satisfying; somewhat satisfying;
moderately satisfying; highly satisfying; extremely satisfying)
Self-Identity How related was consuming FOOD GROUP to your self-identity the past 2 .12
days?
(Continuous 0 to 10) (Not at all related; somewhat related; moderately
related; highly related; extremely related)
Notes: Dependent variables (DV) and predictors, together with their corresponding evaluative
question. Shown underneath each question is the scoring range and corresponding labelling. The
ICC2 establishes participants’ agreement on their judgments for an evaluative question across the 16
food groups in Table 1. Because the ICC2 estimates random effects, these values can be generalized
to other individuals from the same population. The two DVs, frequency and portion, were combined by
calculating their product to create an overall measure of consumption. For each judgment question,
FOOD GROUP refers to 1 of the 16 food groups assessed in Table 1.

Participants were informed that we were interested in their eating and covid

stress over a two-week period, without disclosing our exact research questions.

Then, without participants knowing, we randomly assigned them to one of the two

groups, “the eating group” or “the covid stress group”. The covid-stress group was a
 16

follow-up to Pedersen et al. (2022) and will be reported elsewhere. During the first

and last timepoints, both groups filled out the exact same surveys, differing only on

whether the eating survey came first (eating group) or after the covid stress survey

(covid stress group). The main difference between the two groups were in timepoints

two, three and four, when the eating group only assessed their eating behaviour and

the covid stress group only assessed their covid stress experience.

Both groups filled out the same questionnaires in differing orders on timepoints

one and five, plus additional questions at the end that collected demographic

information and individual difference measures (age, gender, education, household,

religious affiliation, current dieting status, diet restrictions, food responsibility in the

household, past eating disorders, covid related measures and survey experience). In

this article, only the 70 previously described participants from the eating group and

their eating behaviour results are reported.

Results
All following data analyses reported next were conducted using R (R Core

Team, 2019) and R Studio version R-3.6.1 (RStudio Team, 2020) and all data and

analysis scripts for this study will be made available online.

Consumption measure

As described earlier, participants judged how frequently they consumed each of

the 16 food groups at each of the 5 time points and also estimated the typical portion

size (in handfuls) consumed across those occasions. We then combined these two

measures for each food group at each time point by taking their product, thereby

establishing the overall amount consumed. All following analyses were performed on

this combined measure of consumption, except where noted otherwise. Interested

 17

researchers can access the raw data online if they wish to analyse consumption

frequency and/or typical portion size separately.

Implicit learning effects

Of interest in the first analysis was to assess whether the same implicit learning

effects, observed previously for stress in Peterson et al. (2022), would also occur for

food consumption here. Across time points, did the correlation between consumption

and each consumption motive increase? To assess this issue, we computed the

difference between timepoint one and timepoint five in the correlation between

consumption and each consumption motive.

Specifically, we used the raw judgments at time point one to compute the

correlation between consumption and each consumption motive across the 16 foods

for each participant, resulting in 6 correlations for each of the 70 participants. We

then repeated this procedure for time point 5, again resulting in 6 correlations for

each of the 70 participants. For each consumption motive, we then subtracted its

correlation at timepoint one from its correlation at time point five. A positive value

indicated that the strength of correlation increased across time points, implying an

implicit learning effect. In contrast a negative value indicated a decrease in

correlation strength across timepoints.

Figure 1 presents the results for each consumption motive, with each dot

representing the increase in correlation strength for a single participant from

timepoint 1 to timepoint 5 for a single motive. Although, large individual differences

emerged, overall, the median difference for each consumption motive was positive,

suggesting the presence of implicit learning across timepoints.

 18

Figure 1
Individual learning effects

Note: Violin-boxplot represents the distribution of how much the correlation in timepoint 1 and
timepoint 5 differed from each other for each participant. Each dot represents the difference for a
single participant

To assess these possible effects statistically, we used one-sample t-tests to

determine whether the differences in correlations for each consumption motive were

significantly larger than 0, indicating that the strength of correlation had increased

from timepoint one to timepoint five across participants. Because the differences for

some consumptions violated the normality assumption, we used Wilcoxon signed

rank tests. For one-tailed tests that predicted an increase in correlation strength from

timepoint 1 to timepoint 5, these increases were significantly greater than 0 for

healthiness (p = .008), self-identify (p = .035), and fillingness (p = .002). Although the

 19

increases were in the predicted direction for taste, automaticity and emotional

satisfaction, they did not reach significance (for enjoyment of taste, p = .125, for

automaticity, p = .146; for emotional satisfaction, p = .055).

Lastly, for each participant, we computed a combined measure of the

difference in correlation strength between timepoint one and five across all six

consumption motives together. Specifically, we computed the sum of the six

differences in correlation for each participant (shown in Figure 1), resulting in one

value for each of the 70 participants (median = .47, IQR =1.25 range= -1.85 to 1.84).

Analogous to the significance analysis for the individual consumption motives, we

performed a Wilcoxon signed rank test, which found that the mean of the combined

correlation difference score was significantly greater than 0 (p= .001).

In summary, we found large individual differences in how the strength of

association between consumption and consumption motives changed from timepoint

one to timepoint five. Nevertheless, three of our six consumption motives exhibited

significant evidence for implicit learning. Furthermore, when all six consumption

motives were combined, a highly significant learning effect emerged across

participants. Analogous to Pedersen et al. (2022), when participants evaluated their

consumption motives across timepoints, correlations between these motives and

consumption increased. Possible explanations of this implicit learning effect are

addressed later.

G Theory and ICC results

Of interest in this next analysis was assessing the relative importance of

different sources of variance in our data related to consumption, including food

groups, participants, timepoints, their interactions, and residuals. Of particular

 20

interest was assessing the stability of consumption for the 16 food groups across the

5 timepoints.

Generalizability Theory (“G theory”) offers a method for both breaking out

important sources of variance in consumption and for assessing its stability

(reliability). Another way of stating this is that G Theory allows one to distinguish how

much variance in a measure, such as consumption, is explained by the different

random effects (Monteiro et al., 2019). In our study, the random effects included food

groups, participants, timepoints, and all the two-way interactions between them

(participant x food group; participant x timepoint; food group x timepoint). A final

variance component for the residuals was also estimated.

We first assessed the relative sizes of these variance components. As Panel

A in Figure 2 illustrates, large differences emerged between them. As can be seen,

the largest variance component was for the interaction between participants and food

groups, followed by the variance components for food groups and participants. In

other words, the variability in consumption observed at the group level largely

reflected systematic differences across the 16 food groups, across the 70

participants, and, most significantly, in the interaction between food groups and

participants. Surprisingly, timepoint barely explained any variance in consumption

(nor did any of its interactions), providing a first indication of how stable the ratings

remained across the five timepoints.

In terms of consumption, these results first indicate that consumption hardly

varied across timepoint. Consumption somewhat varied across individuals, with

some individuals consuming more than others. Consumption varied much more

across food groups, with different food groups (not surprisingly) being consumed

much more often than others (as illustrated in detail later).

 21

Figure 2
G-Theory Results for original data and residuals from a simple linear regression

Note: Results from the G-Theory analysis. Panel A shows the variance in consumption from each of
the possible sources when explaining the raw original data. Panel B shows the same results when
explaining the residuals of a simple linear regression analysis performed on the group data.

Perhaps most notably, the largest variance component was for the interaction

between participants and food groups, indicating tremendous individual differences in

how much different participants consumed different food groups (also illustrated and

quantified in detail later).

As we just saw, consumption appeared to be remarkably stable across time

points. To examine this finding more closely, we next computed intraclass correlation

coefficients (ICCs) for individual participants that assessed how stable their

consumption estimates were across time points. We also computed ICCs for each of

the six consumption motives as well, to examine how stable these judgments were

across time points.

 22

In general, ICCs provide researchers with a means of assessing how much

different judges agree in their judgements for a set of judged objects (Koo & Li, 2016;

Shrout & Fleiss, 1979). Applying this general framework here, the 5 timepoints for a

single participant served as judges and the 16 food groups served as the judged

objects. For a given measure, such as consumption, the resulting ICC estimated

how much the 5 time points agreed in an individual’s judgments for the 16 food

groups. In other words, the ICC assessed the stability of an individual’s judgments

for the 16 food groups across the 5 time points. To generalise whether these ICCs

for the 5 timepoints assessed here to larger population of time points for each

individual, we used the random effects version of the ICC, the ICC2 (Shrout & Fleiss,

1979).

For each participant, we computed a total of nine ICC2s: Three ICCs related

to the dependent variable for consumption (frequency, typical portion size, and their

product); six ICCs for the six motives used to predict consumption. ICC2s

approaching 1 can be interpretated as illustrating high stability across the five time

points (i.e., participants rated the 16 food groups consistently across timepoints).

Conversely, ICC2s approaching 0 imply no stability across time points (i.e.,

participants rated the different food groups with no consistency across the

timepoints).

Figure 3 presents the results. Consistent with the findings from G Theory in

Figure 2A, we observed high stability across time points for the majority of

participants. Overall, the highest agreement was observed for healthiness (median

ICC2 = .84, IQR = .13), followed by self-identity (median ICC2 = .76, IQR = .31),

emotional satisfaction (median ICC2 = .76, IQR = .20), and automaticity (median

ICC2= .75, IQR = .15). Agreement was somewhat lower across timepoints for the
 23

three dependent measures, although still remaining quite high: consumption (median

ICC2 = .59, IQR=.20), frequency (median ICC2 = .67, IQR = .14), typical portion size

(median ICC2 = .53, IQR = .15). The overall median ICC2 across all measures

(dependent variables and predictors) and individuals was .70 (IQR = .22).

Interestingly, agreement for typical portion size varied the most, suggesting that

individuals varied somewhat in the portion sizes consumed on different occasions.

In summary, both the G-Theory and ICC2 results demonstrated a surprising

lack of intraindividual variability across the five time points, disconfirming our original

prediction that variability would be higher. In other words, participants were

remarkably stable in how they evaluated the 16 food groups across the 5 timepoints

for both the dependent variables related to consumption and for the six predictors for

consumption motivation. One implication of this result is that it may typically be

sufficient to assess an individual’s consumption for a single time point, rather than at

multiple time points, unless implicit learning is of interest.

It’s perhaps worth noting in this regard that the implicit learning effects noted in

the previous section are not inconsistent with the stability of consumption motives

observed here. Implicit learning, for example, could have occurred across patterns of

consumption motives that remained relatively stable across time points. Although the

relative importance of the six consumption motives remained relatively constant for

an individual across timepoints, their correlations with consumption could have

simultaneously increased.
 24

Figure 3
Distribution of individual ICC2 values representing stability of each rating across
timepoints

Note: Violin-Boxplots showing the distribution of individual ICC2 values for each rating question.
Separately for each participant the agreement of how they rated the sixteen food groups was
assessed for each rating question across the five timepoints (5 timepoints were used as judges and 16
food groups as judged objects). Each dot represents the ICC2 of one participant for one rating
question.

Based on these results, we removed timepoint from all later analyses. Doing

so greatly simplified the results reported and allowed us to focus on other patterns in

the data. Specifically, to simplify the data set for each of the 70 participants, we

computed their average judgment across timepoints for each of the 16 food groups,

once for consumption and once for each of the six consumption motives. Doing so

further removed the small variant components for timepoint as well as their
 25

interactions observed in the G-Theory results, leaving only food group and

participants as the remaining random effects. Researchers interested in examining

the analyses to follow on individual time points can do so by accessing the complete

raw data set online.

Variability for individuals and food groups

As we just saw, little variability occurred for timepoints, with consumption

judgments remaining remarkable consistent across them. Here we next look at two

further variance components in Figure 2A for which there was much larger variability:

participants and food groups.

Figures 4A and 4B visualise the variability for participants and food groups.

Figure 4A plots participant means for consumption across timepoints for each food

group. Each of the six panels in Figure 4B similarly plots participant means for one of

the six consumption motives across timepoints for each food group.

As Figure 4A illustrates, consumption varied greatly across food groups and

participants (0 to a maximum of 26.5 handfuls consumed), with a median value of

1.24 handfuls (IQR = 2.92). Specifically, median consumption varied widely between

food groups, with participants varying widely within food groups as well. These

findings further illustrate the large variance components seen earlier for food groups

and participants in Figure 2A. The highest median consumption, along with the

highest variability across participants, emerged for vegetables (median = 5.20, IQR =

4.01), bread (median = 3.54, IQR = 3.08), dairy and egg (median = 3.00, IQR = 3.50).

and fruit (median = 2.90 handful, IQR = 3.00).

Figure 4B presents analogous results for the six consumption motives. Again,

large amounts of variance emerged across both food groups and participants.
 26

Variability across participants tended to be especially high for self-identity,

automaticity, and emotional satisfaction across food groups, demonstrating large

Figure 4A
Distribution of participants’ average consumption for each food group

Notes: Violin-boxplots show the distribution of how many handful participants reported to have
consumed. Each participant’s consumption was calculated by taking their average across the five
timepoints. Each dot represents the average consumption for a single participant.
 27

Figure 4B

Distribution of participants’ average predictor rating for each food group

Notes: Violin-boxplots show the distribution of how participants rated each food group on the six
predictors. Each participant’s score was calculated by taking their average score across the five
timepoints. Each dot represents the average score for one participant.
 28

individual differences on these measures. Even for the measures commonly

assumed to be more objective such as healthiness and fillingness, large individual

differences emerged, even though some agreement emerged for foods considered

healthy (i.e., lower variability across individuals for fruit, vegetables, fish and seafood)

and for foods considered unhealthy (sweets, salty snack, fried food). When

comparing food groups across consumption motives, fruit and vegetables were

consistently judged high on all predictors, whereas plant-based foods were

consistently judged low ( with the exception of healthiness).

The individual by food group interaction for consumption

As we saw earlier in Figure 2A, the largest variance component for

consumption was the participant by food group interaction. Figure 5 visualises this

variance component. Specifically, each row visualises 1 of the 70 participant’s

consumption for the 16 food groups shown across the 16 columns. As a cell

becomes darker, the participant consumed more of the food group (averaged across

the five time points). As a cell becomes lighter, the participant consumed less of the

food group. As the legend on the upper left illustrates, the unit of consumption is the

number of handfuls consumed over the past two days, ranging from 0 to more than

21. To increase the legibility of this figure, discrete bins of handfuls were plotted

instead of the continuous measure (e.g., 11 < handfuls < 16).

Figure 5 visualizes the variance components for both food groups and

participants. For food groups, the darkest columns indicate the foods consumed

most often. For participants, the darkest rows indicate the participants who

consumed the most food overall. Hierarchical clustering of the columns clusters food

groups consumed similarly across individuals. Similarly hierarchical clustering of the

rows clusters individuals who consumed the food groups similarly.

 29

Figure 5
For consumption, the Interaction between participants and food groups.

Notes: Heatmap shows consumption (product of consumption frequency and portion size) in handfuls
that each participant consumed on average (across the five timepoints) within two days. Each row
represents one participant’s values for each food group. Each column shows the consumption across
participants for one food group. For simplification the exact number of handfuls was further coded as
0, between 0 and 1, between 1 and 3, between 3 and 6, between 6 and 11, between 11 and 16,
between 16 and 21 and above 21). White tiles represent 0 handfuls consumed (no consumption). The
darker the tiles the greater the amount of handful consumed.
 30

Most importantly, Figure 5 visualises the participant by food group interaction

as differences between the rows for individuals. If all individuals had exhibited the

same pattern of consumption across the 16 food groups, all 70 rows should look

identical. To the extent that the pattern of consumption across columns varies for

different individuals, this reflects the individual by food group interaction. As the

clusters of participants along the left of Figure 5 illustrate, participants differed

considerably in their patterns of consumption across food groups.

The ICC2 can be used to quantify this interaction. In this context, the judges

are the 70 participants, and the judged objects are the 16 food categories. Of

interest is how much the participants (the judges) agree in their consumption of the

16 food categories (the judged objects). As Table 2 illustrates, the ICC2 for

consumption was only .30, indicating that, on average, one participant’s consumption

of the 16 food categories only correlated .30 with another participant’s consumption

of them. This low value for this ICC2 explains why the largest variance component in

Figure 2A was for the participant by food group interaction.

Table 2 also provides ICC2 values for the six consumption motives,

establishing how much participants agreed in these judgments across the 16 food

categories. As can be seen, the agreement tended to be low across all 6

consumption motives (median= .24, IQR= .09), although agreement was relatively

high for judgments of healthiness, as it often is (ICC2 = .62; see Werner et al., 2022).

In summary, we observed large differences between the 16 food groups, along

with large individual differences in how the 70 participants evaluated them. These

results provide evidence that using the 16 food groups (instead of a much larger

number of individual foods) still allowed us to capture large individual differences in

both food groups and participants.

 31

Individual correlation and regression analys es

We next explored how well the six consumption motives predicted

consumption for participants. In this analysis, we again used each participant’s

average scores across time points for both consumption and the consumption

motives. For each participant, we computed the correlation between consumption

and each consumption motive across the 16 food groups, resulting in 6 correlations.

Figure 6 presents the distribution of correlations for each consumption motive, with

each point being a correlation for one participant. Correlations approaching 1 or -1

indicate that a consumption motive predicts consumption highly, whereas correlations

approaching 0 indicate a lack of prediction.

Figure 7 presents the same results as Figure 6 but as a heatmap that

visualises each participant’s prediction profile. Specifically, each row visualises the

six predictive correlations for a single participant, with increasingly red cells

representing increasingly positive correlations and increasingly blue cells

representing increasingly negative correlations.

 32

Figure 6
Results of the individual correlation analysis showing general trends across the group

Notes: Violin-boxplots present the results of the individual correlation analysis. Correlations were
performed on the average participant scores across the five timepoints. For each of the 70 participants
the correlation between consumption and the six predictors across the sixteen food groups were
computed. Each dot represents the correlation of one single participant.

Hierarchical clusters on the left capture groups of participants with similar

prediction profiles; clusters along the top capture similarity in prediction across

consumption motives. Across the majority of participants, automaticity most strongly

predicted consumption, followed by self-identity and enjoyment of taste (Figures 6

and 7). Healthiness and fillingness, in contrast, exhibited the weakest correlations

with consumption across participants. Finally, all six predictors exhibited

considerable variability across participants, indicating large individual differences in

consumption motives. Correlations of healthiness, for instance, ranged from highly

negative to highly positive across individuals. Similarly, fillingness ranged from

weakly negative to highly positive.

 33

How much variance do the six consumption motives explain in an individual’s

consumption? To assess this question, we regressed an individual’s consumption

scores onto their judgments for the 6 motives across the 16 food groups.

Again, these analyses used mean judgments across the five time points. All

measures were standardized and entered into a simple linear regression for each

participant, with no random effects or interactions. Across these 70 individual

regressions, the median R2 was .71, indicating that the 6 consumption motives did an

excellent job of explaining consumption variance in individual participants (IQR= .15,

range= .36 to .98). As in Werner et al. (2022), we sampled a relatively

comprehensive set of consumptions motives that was sufficient to explain

consumption well for most participants.

One final analysis further demonstrates how well the six consumption motives

explain the consumption variance in this study. We performed a simple linear

regression at the group level on the original data set assessed with G Theory in

Figure 2A. Specifically, we simultaneously regressed consumption for each

 34

Figure 7
Results of the individual correlation analysis visualising each participant’s predictive
profile for consumption.

Note: Heatmap presenting the same results as Figure 6. Each row represents the correlation profile
of one participant. Each column shows all participants’ correlations for one predictor. The colour of
each tile reflects the correlation result. The more a correlation approaches 1 the redder the cell, the
more it approaches zero the whiter the cell and the more the correlation approaches -1 the bluer the
cell.
 35

participant for each food group for each timepoint onto the six consumption

motives, again for each participant for each food for each timepoint. No random

effects or interactions were included.

After performing this regression, we obtained the residuals that remained after

predicting consumption with the six consumption motives and submitted them to a

second G Theory analysis. Figure 2B presents the results. Of interest was how

much the six consumption motives were able to explain the original variance

components in Figure 2A. If the six consumption motives are sufficient to explain the

original variance components for participants, foods, participants X foods, and the

original residuals, then all these variance components should decrease significantly

after regressing consumption onto the six consumption motives. Figure 2B

illustrates that this was indeed the case. By in large, all the original variance

components almost disappeared entirely, indicating that the six consumption motives

captured most of the variance in them. These results further demonstrate that the six

consumption motives are sufficient for comprehensively explaining the variance in

consumption for participants, foods, and the participants by foods interaction.

Individual self-assessment
Lastly, we addressed the question of how much insight participants have into

what predicts their consumption (as measured with their judgments of consumption

and the six consumption motives: Figure 6 and 7). To do so, we asked participants

at the end of each session to judge how much they believed each of the six

predictors had influenced their food consumption the past two days.

We again assessed the stability in the self-assessment ratings across the five

time points (i.e.., how much did participants self-assessment judgments vary across

timepoints). Analogous to earlier assessments of stability, we used the ICC2 to

 36

quantify how much each participant agreed on average with their own self-

assessment judgments across the five timepoints. In other words, the five timepoints

were treated as judges and the six consumption motives were treated as judged

objects. Overall, participants ratings were again quite stable across timepoints

(median= .64, IQR= .35) though the level of stability differed greatly across

participants (range = .03 to .94). Stability here was slightly lower than observed

earlier for participants’ consumption and consumption (median= .70, IQR= .22).

Again, though, their judgments showed considerable stability across time points.

Taking into account the considerable stability across timepoints, we again decided to

compute the average for each rating across the five timepoints separately for each

participant and use these averages in all later analyses.

Figure 8
Each participant’s averaged self-rated importance of each of consumption motive.

Note: Violin-boxplots present participants’ average self-assessed importance of each of the six
predictors. Participants rated importance on a continuous scale from 0 to 10. Each dot represents one
single participant’s rating. Mirroring our previous results, we observed large individual differences in
what participants believed to have influenced their food consumption
 37

Across participants, taste emerged as the consumption motive that they

believed had most influenced their consumption, followed by emotional satisfaction,

automaticity, and fillingness. In contrast, self-identity was rated the least influential.

These results diverge considerably from the pattern of prediction in their SAM2

judgments of consumption motives (Figure 6 and 7). Although enjoyment of taste did

appeared as a strong predictor in those judgments, automaticity emerged as a much

stronger predictor. The largest difference emerged for self-identity. Although it

emerged as highly predictive of consumption in participants’ SAM2 judgments

(Figures 6 and 7), participant did not believe intuitively that it had influenced their

consumption behaviour recently (Figure 8).

Finally, we assessed how much the SAM2 judgments of consumption motives

matched the self-assessed motives that participants believed had influenced their

recent consumption. Based on previous studies, we expected to see little agreement

between the implicit and explicit measures. To enable a comparison for each

participant, we took their correlations for the 6 consumption motives in Figure 7 and

correlated them with their average self-assessment judgments in Figure 8. If a

participant had insight into the consumption motives that influence their consumption,

then these correlations should be high. Motives that participants believe influence

their consumption should be highly correlated with consumption in their SAM2 data,

whereas motives they believe are uninfluential should be exhibit weak correlations.

Although the SAM2 judgments only capture correlations between consumption and

the consumption motives, if a participant’s explicit self-assessments of what

influenced their consumption are correct, then the correlational SAM2 data should be

consistent with their causal judgments.

 38

Figure 9 presents the results of this analysis. As can be seen, these

correlations tend to be low, with a median of .0.02 (IQR = .8). In general,

participants had little insights into their consumption motivation associated with their

eating behaviour. Although a subgroup of participants demonstrated moderate to

high level insight into their consumption motivations, another subgroup demonstrated

the opposite. It is difficult to tell from these data whether some participants actually

have high insight, or whether they’re simply on the upper end of random variability.

Individual study experience

Lastly, we were interested in gaining a better understanding of how

participants experienced taking part in our study and completing our eating motive

questionnaire. To assess this, we asked participants at the end of timepoint five to

judge their experience on three rating questions using a continuous scale from zero

to 10.

Figure 9
Correlation between SAM2 prediction and participant’s self-assessment

Note: Figure shows the correlations between what we predicted to be important for each participant’s
consumption and the participant’s self-assessment. Each dot represents the correlation for one single
participant.
 39

First, we asked participants whether they had learned something

interesting/useful about their eating behaviour from participating in the study (M=

6.02, SD = 2.62). Second, we asked whether participants felt they learned

something about the factors that influence their eating behaviour (M=5.61, SD =

2.64). Third, we asked whether they had learned to better regulate their eating

behaviour (M=4.26, SD = 2.74). Overall, from these responses, it appears that most

participants felt that taking part in our study resulted in increased knowledge about

themselves. Although as discussed previously, this was not reflected in changes to

their self-assessment. Interestingly, about 10% of participants reported, unprompted

in the open feedback section, that they and enjoyed the study and that it had helped

them with their eating behaviour (e.g., “The questions regarding my feelings and

eating really made me think about what I eat;” “You've made me keep a better note of

the food I eat & I'm more aware of my reasons for eating at times.”)

Discussion

In this study, we assessed individuals’ motivation for consuming sixteen food

groups over a two-week period, exploring implicit learning effects as well as individual

differences in what predicts an individual’s consumption and the stability of these

motives across timepoints. We also assessed participants’ insights into their eating

motives.

Implicit learning effects

Comparing the difference in correlation strength between timepoint one and

five for each of the six consumption motives revealed large individual differences in

change across time-points. Nevertheless, implicit learning effects emerged (Figure

1). Significant implicit learning effects occurred for three of the six measures
 40

(healthiness, self-identity and filligness), and also for the combined measure

(summed up differences in correlation across all 6 measures) of difference.

As noted earlier, Pedersen et al. (2022) observed similar implicit learning for

eustress and distress (i.e., correlations with stress predictors increased over time).

The attenuation of correlation offers a potential explanation for all these effects

(Schmidt & Hunter, 1996). As is well known, when a true correlation in nature is

measured, the observed correlation is attenuated by the reliability of the two

measures being correlated. As these the reliabilities of these measures decrease,

the observed correlation is attenuated, relative to the true correlation in nature.

The implicit learning effects that Pedersen et al. (2022) observed and that we

observed may reflect a decrease in attenuation across timepoints. The observed

correlations at each timepoint can be described as dependent on the correlation of

the ‘true scores’ and the reliability of the measured variables (Schmidt & Hunter,

1996). By asking participants to repeat the measures five times, their measurement

reliability might have increased. This effect might have been especially pronounced

for portion size. Although participants might have initially been unfamiliar with

measuring their foods in handfuls and might have consequently been not very

reliable in their estimates during timepoint one (Gibson et al., 2016), knowing they will

have to report their portion size again in the subsequent timepoints might have made

them more aware of their portion size, thereby increasing their accuracy at later time

points. Similar effects might have occurred for the predictors. Across measures, as

their reliability increased over time, correlations between them may have increased.

Although more research is needed, this account offers one way to understand the

implicit learning effects observed here and in Pedersen et al. (2022).

Individual differences in consumption motives

 41

Individual correlation analysis established large individual differences in

consumption motives, although automaticity emerged as consistently high in

importance across participants (Figures 6 and 7). Additionally, the six consumption

motives explained large amounts of variance in individual consumption (median= .71,

IQR= .15). Even though large individual differences emerged in consumption

motives, G Theory and ICC2 results revealed that an individual’s motives tended to

remain highly stable across five timepoints over a two-week period (Figures 2 and 3).

Lastly, we observed little agreement (Figure 9) between a participant’s predictive

profile in their SAM2 judgments of consumption motives (Figure 5 and 6) and their

explicit self-assessment of what they believe influences their food consumption

(Figure 8).

Our findings furthermore insight into the consumption behaviour of UK

residents. We found large individual differences in participants patterns despite

assessing the consumption of relatively broad food groups (instead of assessing

individual foods). As Figure 3 illustrated, peoples’ diets differed greatly from each

other in the food groups they consumed most often (Figure 5) and how they

perceived the 16 food groups with respect to the consumption motives (Figure 4).

The only agreement that emerged was that most (but not all) participants appeared to

consume fruit and vegetables the most and agreed that they are relatively high in

healthiness. Plant based products appeared to be rarely eaten by most of the

participants. Both of the plant-based food groups were rated the lowest on self-

identity, automaticity, and taste, despite being judged as especially healthy food

groups.

These particular results are in line with previous findings in Danish consumers,

where taste was a barrier for increasing plant-based food consumption in meat eaters
 42

(Reipurth et al., 2019). Even though some questions about the healthiness and

sustainability of plant based alternatives remain (Choudhury et al., 2020), their

benefit compared to animal agriculture is evident (Mäkiniemi & Vainio, 2014) and

they play an important role in making diets more sustainable to effectively address

climate change (Sabaté & Soret, 2014). Our results point towards possible new

starting points for interventions, such as focusing on changing people’s perception of

plant-based products’ taste and increasing their association with people’s self-

identity.

In line with previous findings (Werner et al., 2022), automaticity, emerged as

the most important motive for consumption of the 16 food groups across participants.

This result mirrored other findings such as Brug et al. (2006) and Riet et al. (2011)

who both found habit to be most predictive for consumption. In contrast participants

differed greatly on as how predictive the other five motives were for their

consumption. Furthermore, the same pattern was not reflected in participants’ self-

assessments, who, with some exceptions, judged taste to be most influential for their

food consumption. Similar we found healthiness to be rated as highly influential for

consumption across participants, whereas the same importance did not emerge from

the SAM2 data for the majority of participants.

The observed stability of consumption motivation across time matched

previous findings showing the stability of eating motivations across eating situations

(Werner et al., 2022). These results might be again reflective of the aforementioned

importance of habit for predicting consumption – since habitual consumption would

be expected to emerge as consistently important over time. These findings also

indicate that, in future research, a single assessment at one timepoint would be

enough to reliably establish individual consumption motives.

 43

Self-assessments of eating motives

In general, we found participants’ explicit self-assessments of how much the

six consumption motives influenced their consumption (Figure 8) were largely

unrelated to the most predictive consumption motives in their SAM2 data (Figures 5

and 6). In other words, participants do not have much insight into what actually

motivates their behaviour. This might further provide an explanation for the difficulty

people encounter when trying to change their eating behaviour or maintain weight

loss (Wing & Phelan, 2005), or when they try to establish a more sustainable diet.

Because of this poor insight, individuals may often adopt ineffective approaches to

changing their diet. If participants received more accurate accounts of their eating

motives, such as from a SAM2 analysis, perhaps they would be more successful in

altering their diets. If, for example, an individual mistakenly believes that healthiness

motivates their consumption when, in actuality, automaticity does, focusing on

healthiness to change a diet may be ineffective, whereas focusing on automaticity

and habits may be more effective.

The survey experience questions in timepoint five, along with the open

comments, pointed towards the possibility of using a SAM2 approach to help people

gain a better understanding of what drives their food consumption, potentially even

using it as a brief intervention by providing participants with feedback about their

consumption motives. Further research, however, is needed to systematically

assess whether SAM2 may be useful in helping change people’s eating behaviour.

Strengths and limitations

A strength of the present study was the inclusion of multiple timepoints,

allowing us to investigate the stability of consumption motives across a two-week

period. Further research into temporal stability would be needed to assess whether
 44

longer-term stability exists over month and years, and to understand how motives

possibly evolve over time.

A second strength of this study was to situate all measures in the past two

days, allowing us to assess potential differences across time. A further strength of

our study was additionally recording participants self-assessment, allowing us to

compare those to our SAM2 measure and providing an insight into participants own

beliefs as to what motivates their consumption.

One limitation of our study, due to considerable time constraints, was to not

ask participants to judge the food groups within specific eating situations (e.g.,

sweets for breakfast). Although our previous work suggested that eating motives

remain remarkably stable across eating situations (Werner et al., 2022), other

research has shown differences in motives across situations (Cadario & Morewedge,

2022). In future studies it would be useful to include diverse eating situations and

assess their time-stability.

A further limitation might have been the use of the broad categories of food

groups. Although the sixteen chosen food groups appeared to capture individuals’

diets well and generated large individual differences in judgments of consumption

motivation, in the future, it would be useful to additionally record the specific foods

participants consumed within these food groups. Similarly, while our six predictors

explaining consumption well for most participants, there were some participants

whose consumption our motives appeared to not explain well. In future studies, it

would therefore be useful to use a multi-method approach that includes qualitative

interviews to help identify which additional motives might be relevant for some

participants.
 45

The greatest limitation of our study was our reliance on self-reported,

retroactively collected data, forcing participants to remember their consumption and

consumption motives the past two days instead of recording them in the moment.

Thus, the judgments we collected might be prone to various kind of biases (e.g.,

recency effects). Additionally, these judgments might potentially tap into more

general judgments about consumption and motives instead of actual judgments at

specific timepoints. An account along these lines might offer an alternative

explanation of why we found little variation across timepoints. In the future it could be

beneficial to use ecologic momentary assessment methods to assess consumption

motivation before and/or straight after eating occurs, thereby gaining more accurate

measures.

Furthermore, even though we provided participants with brief explanations for

each predictor, it would be beneficial in future studies to assess, especially for the

more complex predictors such as self-identity and automaticity, how exactly

participants made their judgments. A possible method might be to include qualitative

interviews in addition to our quantitative approach, asking participants to verbalise

their thought and judgment processes.

Lastly, we collected our data in early 2021 while the UK was in a nationwide

lockdown due to the COVID pandemic. A replication of the study in the future would

allow to assess whether consumption motivation had been affected by the pandemic

e.g., by participants spending more time in their homes.

Conclusion

In summary we observed evidence for implicit learning effects across

timepoints, potentially indicating an increase in reliability of measures across time.

Furthermore, we found consumption motivation across participants to be remarkably

 46

stable across time. Our six predictors were able to explain large amounts of variance

on the individual participant level for the majority of participants. We found large

individual differences between what predicts consumption in individuals, although

automaticity emerged as a strong predictor across participants. Lastly, we found that

most participants had little insight into the motives associated with their consumption.

Declarations
Funding: This work was funded by the Institute of Neuroscience and Psychology at

the University Glasgow (58 Hillhead Street, Glasgow G12 8QB).

Conflicts of interest/Competing interests: The authors declare no conflict of

interests/competing interests.

Availability of data and material: Please find the data and all analysis scripts on the

Open Science Framework (OSF https://osf.io/4fcw7/)

Ethics: This project was approved by the University of Glasgow College of Science &

Engineering Ethics Committee (Date: 07/01/2021, no: 300200092).

 47

References
Al-Ghussain, L. (2019). Global warming: Review on driving forces and mitigation.

Environmental Progress & Sustainable Energy, 38(1), 13–21.

https://doi.org/10.1002/ep.13041

Barsalou, L. W. (2020). Challenges and Opportunities for Grounding Cognition. Journal of

Cognition, 3(1), Article 1. https://doi.org/10.5334/joc.116

Birt, C., Buzeti, T., Grosso, G., Justesen, L., Lachat, C., Lafranconi, A., Mertanen, E.,

Rangelov, N., & Sarlio-Lähteenkorva, S. (2017). Healthy and sustainable diets for

European countries. http://hdl.handle.net/1854/LU-8521128

Bisogni, C. A., Connors, M., Devine, C. M., & Sobal, J. (2002). Who We Are and How We

Eat: A Qualitative Study of Identities in Food Choice. Journal of Nutrition Education

and Behavior, 34(3), 128–139. https://doi.org/10.1016/S1499-4046(06)60082-1

Brown, R., Gray, A. R., Chua, M. G., Ware, L., Chisholm, A., & Tey, S. L. (2021). Is a Handful

an Effective Way to Guide Nut Recommendations? International Journal of

Environmental Research and Public Health, 18(15), Article 15.

https://doi.org/10.3390/ijerph18157812

Brug, J., de Vet, E., de Nooijer, J., & Verplanken, B. (2006). Predicting Fruit Consumption:

Cognitions, Intention, and Habits. Journal of Nutrition Education and Behavior, 38(2),

73–81. https://doi.org/10.1016/j.jneb.2005.11.027

Cadario, R., & Morewedge, C. K. (2022). Why do people eat the same breakfast every day?

Goals and circadian rhythms of variety seeking in meals. Appetite, 168, 105716.

https://doi.org/10.1016/j.appet.2021.105716

Calle, E. E., & Kaaks, R. (2004). Overweight, obesity and cancer: Epidemiological evidence

and proposed mechanisms. Nature Reviews Cancer, 4(8), Article 8.

https://doi.org/10.1038/nrc1408

Choudhury, D., Singh, S., Seah, J. S. H., Yeo, D. C. L., & Tan, L. P. (2020).

Commercialization of Plant-Based Meat Alternatives. Trends in Plant Science, 25(11),

1055–1058. https://doi.org/10.1016/j.tplants.2020.08.006
 48

Cousino Klein, L., Corwin, E. J., & Ceballos, R. M. (2004). Leptin, hunger, and body weight:

Influence of gender, tobacco smoking, and smoking abstinence. Addictive Behaviors,

29(5), 921–927. https://doi.org/10.1016/j.addbeh.2004.02.023

Dutriaux, L., Clark, N., Papies, E. K., Scheepers, C., & Barsalou, L. (2021). The Situated

Assessment Method (SAM2): Establishing Individual Differences in Habitual Behavior.

PsyArXiv. https://doi.org/10.31234/osf.io/k3mqj

El Ansari, W., Adetunji, H., & Oskrochi, R. (2014). Food and Mental Health: Relationship

between Food and Perceived Stress and Depressive Symptoms among University

Students in the United Kingdom. Central European Journal of Public Health, 22(2),

90–97. https://doi.org/10.21101/cejph.a3941

Gibson, A. A., Hsu, M. S. H., Rangan, A. M., Seimon, R. V., Lee, C. M. Y., Das, A., Finch, C.

H., & Sainsbury, A. (2016). Accuracy of hands v. Household measures as portion size

estimation aids. Journal of Nutritional Science, 5, e29.

https://doi.org/10.1017/jns.2016.22

Groesz, L. M., McCoy, S., Carl, J., Saslow, L., Stewart, J., Adler, N., Laraia, B., & Epel, E.

(2012). What is eating you? Stress and the drive to eat. Appetite, 58(2), 717–721.

https://doi.org/10.1016/j.appet.2011.11.028

Higgins, S. C., Gueorguiev, M., & Korbonits, M. (2007). Ghrelin, the peripheral hunger

hormone. Annals of Medicine, 39(2), 116–136.

https://doi.org/10.1080/07853890601149179

Higgs, S. (2015). Social norms and their influence on eating behaviours. Appetite, 86, 38–44.

https://doi.org/10.1016/j.appet.2014.10.021

Hofmann, W., Vohs, K. D., & Baumeister, R. F. (2018). What people desire,

feel conflicted about, and try to resist in everyday life. In Self-Regulation and Self-

Control. Routledge.

Ignarro, L. J., Balestrieri, M. L., & Napoli, C. (2007). Nutrition, physical activity, and

cardiovascular disease: An update. Cardiovascular Research, 73(2), 326–340.

https://doi.org/10.1016/j.cardiores.2006.06.030
 49

Koo, T. K., & Li, M. Y. (2016). A Guideline of Selecting and Reporting Intraclass Correlation

Coefficients for Reliability Research. Journal of Chiropractic Medicine, 15(2), 155–

163. https://doi.org/10.1016/j.jcm.2016.02.012

Macdiarmid, J. I., Douglas, F., & Campbell, J. (2016). Eating like there’s no tomorrow: Public

awareness of the environmental impact of food and reluctance to eat less meat as

part of a sustainable diet. Appetite, 96, 487–493.

https://doi.org/10.1016/j.appet.2015.10.011

Mäkiniemi, J.-P., & Vainio, A. (2014). Barriers to climate-friendly food choices among young

adults in Finland. Appetite, 74, 12–19. https://doi.org/10.1016/j.appet.2013.11.016

Monteiro, S., Sullivan, G. M., & Chan, T. M. (2019). Generalizability Theory Made Simple(r):

An Introductory Primer to G-Studies. Journal of Graduate Medical Education, 11(4),

365–370. https://doi.org/10.4300/JGME-D-19-00464.1

O’Mara, F. P. (2011). The significance of livestock as a contributor to global greenhouse gas

emissions today and in the near future. Animal Feed Science and Technology, 166–

167, 7–15. https://doi.org/10.1016/j.anifeedsci.2011.04.074

Pedersen, C., Scheepers, C., Werner, J., & Barsalou, L. W. (2022). Learning effects in better

understanding eustress and distress during the covid pandemic. [Manuscript in

Preparation]. School of Psychology and Neuroscience, University of Glasgow.

Reichenberger, J., Schnepper, R., Arend, A.-K., & Blechert, J. (2020). Emotional eating in

healthy individuals and patients with an eating disorder: Evidence from psychometric,

experimental and naturalistic studies. Proceedings of the Nutrition Society, 79(3),

290–299. https://doi.org/10.1017/S0029665120007004

Reipurth, M. F. S., Hørby, L., Gregersen, C. G., Bonke, A., & Perez Cueto, F. J. A. (2019).

Barriers and facilitators towards adopting a more plant-based diet in a sample of

Danish consumers. Food Quality and Preference, 73, 288–292.

https://doi.org/10.1016/j.foodqual.2018.10.012
 50

Riet, J. van’t, Sijtsema, S. J., Dagevos, H., & De Bruijn, G.-J. (2011). The importance of

habits in eating behaviour. An overview and recommendations for future research.

Appetite, 57(3), 585–596. https://doi.org/10.1016/j.appet.2011.07.010

Sabaté, J., & Soret, S. (2014). Sustainability of plant-based diets: Back to the future. The

American Journal of Clinical Nutrition, 100(suppl_1), 476S-482S.

https://doi.org/10.3945/ajcn.113.071522

Schmidt, F. L., & Hunter, J. E. (1996). Measurement error in psychological research:

Lessons from 26 research scenarios. Psychological Methods, 1, 199–223.

https://doi.org/10.1037/1082-989X.1.2.199

Spencer, S. J., Korosi, A., Layé, S., Shukitt-Hale, B., & Barrientos, R. M. (2017). Food for

thought: How nutrition impacts cognition and emotion. Npj Science of Food, 1(1),

Article 1. https://doi.org/10.1038/s41538-017-0008-y

Stroebele-Benschop, N., Dieze, A., & Hilzendegen, C. (2018). Students’ adherence to dietary

recommendations and their food consumption habits. Nutrition and Health, 24(2), 75–

81. https://doi.org/10.1177/0260106018772946

Tobler, C., Visschers, V. H. M., & Siegrist, M. (2011). Eating green. Consumers’ willingness

to adopt ecological food consumption behaviors. Appetite, 57(3), 674–682.

https://doi.org/10.1016/j.appet.2011.08.010

Vainio, A., Niva, M., Jallinoja, P., & Latvala, T. (2016). From beef to beans: Eating motives

and the replacement of animal proteins with plant proteins among Finnish consumers.

Appetite, 106, 92–100. https://doi.org/10.1016/j.appet.2016.03.002

Wehbe, L. H., Banas, K., & Papies, E. K. (2022). It’s Easy to Maintain When the Changes

Are Small: Exploring Environmentally Motivated Dietary Changes From a Self-control

Perspective. Collabra: Psychology, 8(1), 38823.

https://doi.org/10.1525/collabra.38823

Werner, J., Papies, E. K., Gelibter, Elena, & Barsalou, L. W. (2022). Why do you eat?

Establishing Individual consumption motives and their stability across eating

situations.
 51

Willett, W. C. (1994). Diet and Health: What Should We Eat? Science, 264(5158), 532–537.

https://doi.org/10.1126/science.8160011

Wing, R. R., & Phelan, S. (2005). Long-term weight loss maintenance. The American Journal

of Clinical Nutrition, 82(1), 222S-225S. https://doi.org/10.1093/ajcn/82.1.222S

World Health Organization. (2003). Diet, Nutrition, and the Prevention of Chronic Diseases:

Report of a Joint WHO/FAO Expert Consultation. World Health Organization.