Participants were recruited through the online participant recruitment scheme of the [University redacted]. Eligibility requirements meant that participants were aged 18–40, had no history of neurological illness, and no medical history of eating disorders. A total of thirty-eight participants took part in the experiment. Eight participants were excluded from the sample because they scored below-chance in the pairing task (i.e. errors on >50% of trials in at least one of the experimental conditions). One participant had a BMI greater than 30 (obese individual), and was also excluded from the sample given known attentional biases associated with high BMI (Castellanos et al. 2009; Nijs et al. 2010; Yokum et al. 2011). The final sample was 29 participants (13 men) aged 19–35 years (mean = 23.21, SD = 4.126). The sample characteristics are described in Table 1. Participants gave their informed consent prior to the study, with approval by the Ethics Committee, Department of [Redacted].
A set of 60 food pictures was selected from The Foodcast Research Image Database (FRIDa) (Foroni et al. 2013) that included standardised ratings on a number of dimensions such as valence, perceived calories, caloric value per 100 g, familiarity, and discriminability. The food stimuli (530 × 530 pixels) comprised 20 pictures of natural food, 20 pictures of transformed food, and 20 pictures of rotten-food (Fig. 1). These food categories significantly differed in valence (defined in the FRIDa database as the degree to which subjects found the food items pleasant), ranging from ‘very positive’ to ‘very negative’ (natural food, M = 67.09, SD = 11.29; transformed food, M = 64.89, SD = 8.22; rotten food, M = 6.29, SD = 4.90; F(2,38) = 353.486, p < .001), perceived calories (natural food, M = 22.71, SD = 10.91; transformed food, M = 67.32, SD = 12.27; rotten food, M = 25.81, SD = 14.85; F(2,38) = 99.64, p < .001), caloric value per 100 g (natural food, M = 97.35, SD = 105.92; transformed food, M = 298.35, SD = 170.82; rotten food, M = 102.55, SD = 113.18; F(2,38) = 99.64, p < .001), familiarity (natural food, M = 54.71, SD = 19.52; transformed food, M = 51.48, SD = 14.13; rotten food, M = 11.27, SD = 4.89; F(2,38) = 58.210, p < .001), and discriminability (natural food, M = 10.60, SD = 9.49; transformed food, M = 14.73, SD = 10.20; rotten food, M = 42.60, SD = 27.10; F(2,38) = 18.619, p < .001).
Natural and transformed food had comparable valences, and they both significantly differed from rotten food (all t > 22.19, df = 19, all p < .01). Transformed food items had significantly greater perceived calories and caloric value per 100 g than natural and rotten food items (all t > 12.33, df = 19, all ps < .01). Natural and transformed food were comparable in familiarity and discriminability, but both differed from rotten food (all t > 4.30, df = 19, all ps < .01), which by nature is less available in food contexts (i.e. less familiar) and often harder to discriminate than non-rotten food. All food stimuli were comparable for the following variables: size, spatial frequency, brightness, and arousal (all t < 2.44, df = 19, all ps > .05).
Moreover, the stimuli comprised three words “You”, “Friend”, and “Stranger” (font size 18-point, font type Courier New) in black font. Both the labels and the food items were presented centrally on a white background at a distance of 70 cm from the participants’ eyes (monitor resolution 1024 × 786 pixels). E-prime software (Psychology Software Tools) was used to present the stimuli and record behavioural responses.
The experimental session was divided in three runs. Each of the three runs started with the pretest phase, whereby participants were instructed to associate the food items (20 natural, 20 transformed, 20 rotten food items) with one of three things: their self, their best friend, or an unfamiliar person (pretest phase). For example, a participant was told “You are natural food, Tom (the named best friend of the participant) is transformed food and a stranger is represented by rotten food”. Note the meaning of the terms natural, transformed, and rotten food (described above) had been explained to the participants. The food items themselves were not presented at this stage. The pairing food type-label was counterbalanced and pseudo-randomized within participants across runs, and the label ‘you’ was associated in each of the three runs with either natural, transformed, or rotten food.
After this, the matching phase of the experiment was performed. As shown in Fig. 2, each trial started with the presentation of a fixation cross (500 ms), followed by the label (80 ms), and subsequently the food item (duration until response). The food type-label pairing could conform to the instructions given in the pretest phase, or it could be a recombination of a label with a different food item. The pairing of the food item and the label was equally distributed, such that each food item was equally likely to appear with each label, and the ratio between match and mismatch trials was 1:1.
Participants’ task was to judge whether the food item was correctly assigned to the relevant person by pressing one of the two response buttons as quickly and accurately as possible. Response buttons were counterbalanced across participants. Feedback (correct or incorrect) was presented on the screen for 500 ms after participants responded in every trial. Participants were also informed of their overall accuracy at the end of each run.
The overall experiment consisted of 1080 randomized trials presented in 3 runs (360 trials per run) following some practice trials, where self, friend, stranger, and re-paired stimuli occurred equally often in a random order. Thus, there were 60 trials per condition (self-matched natural/transformed/rotten food, self-nonmatched natural/transformed/rotten food, friend-matched natural/transformed/rotten food, friend-nonmatched natural/transformed /rotten food, stranger-matched natural/transformed/rotten food, stranger-nonmatched natural/transformed/rotten food).
After the matching task, to ensure that the participants included in sample were comparable in the these measures that may influence food perception, we measured Body Mass Index (BMI) based on self-reported height and weight, and feelings of hunger by using a 10 cm visual analogical scale (VAS) ranging from 0 (“I am not hungry at all”) to 100 (“I am starving”). In addition, we measured participant’s eating behaviour with the Three-Factor Eating Questionnaire-R21 (TFEQ), which includes three domains of eating behaviour: cognitive restraint (6 items), uncontrolled eating (9 items), and emotional eating (6 items). Most items were rated on a four-point Likert scale ranging from a minimum of 0 to a maximum of 4. One item was rated on an eight-point numerical scale (minimum = 0, maximum = 8) (Cappelleri et al. 2009).
To assess the familiarity effect, we asked participants to report the length of the friendship (in years), level of familiarity on a scale from 1 (not familiar at all) to 7 (highly familiar), the frequency in which they meet on a scale from 1 (not frequently at all) to 7 (very frequently), and whether they meet daily, weekly, monthly, or otherwise.
The results of the self-report measures are reported in the Supplementary material.
Data was analysed using SPSS Statistics 22 (IBM). Analysis were centred on the matching trials, because we were principally interested in testing whether matching a food item with oneself modulates attentional biases towards the food item. Differences in accuracy and reaction times between conditions were analysed using repeated measures ANOVAs with within-subject factors food type (natural, transformed, rotten) and label (you, friend, stranger). To control for the influence of carrying over pairing effects from one block to the next, the repeated measures ANOVAs also included the variable block order (first, second, third, fourth, fifth, sixth) as between-subject factor. Additionally, paired t-tests were performed. Mauchly’s W was computed to check for violations of the sphericity assumption and Greenhouse–Geisser adjustments to the degrees of freedom were applied when needed. The p values were corrected for multiple comparisons using stepwise Bonferroni-Holm correction.
Below we report results on three different measures: participant accuracy, participant response time, and participant responses to our self-report questions.
Table 2 represents mean accuracy for each condition. A three-way repeated-measures ANOVA of accuracy with factors food type (natural, transformed, rotten), label (self, friend, stranger) and block order (first, second, third, fourth, fifth, sixth) revealed the following. First, we found a significant difference between self, friend and stranger as indicated by a main effect of label (F(2,46) = 8.836, p = .002, η2 = .278). The results also showed that the self versus stranger differences were influenced by the order of presentation, as indexed by an interaction between label and block order (F(10,46) = 2.846, p = .014, η2 = .382). Second, we found significant differences between the three food types (i.e. main effect of food type, F(2,46) = 11.815, p < .001, η2 = .339). Importantly, the ANOVA results showed that the food type-label pairing effect was different for the self, the friend and the stranger condition, as indicated by interactions between food type and label (F(4,92) = 5.923, p < .001, η2 = .205) as well as between label, food type, and block order (F(4,92) = 1.985, p = .015, η2 = .301).
To directly test for the pairing effects on the matching processes of the three food types we performed three separate ANOVAs for natural, transformed and rotten food with the within-subject factor label (self, friend, stranger), and the block order as a between-subject factor. For natural food, the results of the ANOVA showed significant differences between self, friend and stranger (F(2,46) = 16.428, p < .001, η2 = .417). Crucially, this effect did not interact with the factor block order showing that the pairing effects on natural food items are independent from the pairing order across blocks. Follow-up t-tests revealed that there was a significant advantage for the process of pairing natural food items with the self over the pairing with the stranger (t(28) = 5.467, p < .000, d = 1.03) and with the friend (t(28) = 2.886, p = .007, r = 0.229). There was also an advantage for pairing natural food with the friend as compared to the stranger (t(28) = 3.119, p = .004, d =. 5894; Fig. 3, see Table 2 for accuracy scores). Furthermore, the significant advantage of the self over the friend label indicates that the pairing effects observed in the self-condition do not reflect familiarity alone.
In contrast, the ANOVA performed for transformed food did not reveal significant differences between labels (F(2,46) = 3.222, p = .049, η2 = .123). For rotten food, the results of the ANOVA showed no main differences between self, friend and stranger (F(2,26) = 2.940; p = .063; η2 = .113), but the order of presentation lead to differences in the association of rotten food with self, friend and stranger (i.e. interaction between label and block order (F(2,46) = 4.642, p < .001, η2 = .502). The follow-up tests showed differences when pairing rotten food with the friend versus the stranger (F(1,23) = 8.844, p = .000, η2 = .658). Although this interaction can be interesting to explain the friend and stranger pairing effects, it expands beyond the aim of the study, i.e. to investigate whether self-association can modulate food attentional bias. Therefore, this interaction will not be further tested. The analysis did not reveal any other main significant effects or in interaction (all Fs < 2.927, df = 1,23, all ps > 0.035).
Participant Response Times
Mirroring the analysis performed for the accuracy measure, we tested differences in reaction times by performing a three-way repeated measures ANOVA with factors food type (natural, transformed, rotten), label (self, friend, stranger), and block order (first, second, third, fourth, fifth, sixth). This analysis revealed significant differences between self, friend and stranger (F(2,46) = 29.013, p < .001, η2 = .558), significant differences between the food types (F(2,46) = 14.579, p < .001, η2 = .388), and significant differences when pairing the three labels – self, friend, stranger, with the three food items – natural, transformed and rotten (F(4,92) = 12.886, p < .001, η2 = .359). In view of the differential effects of associating the three labels with the three food items, we then performed further ANOVAs to directly test for the pairing effects on the three types of food items (natural, transformed, rotten food) separately. The ANOVAs with within-subject factor label (self, friend, stranger) showed a main effect of label for natural food (F(2,56) = 40.445, p < .001, η2 = .591), for transformed food (F(2,56) = 7.021, p = .004, η2 = .200) and for rotten food (F(2,56) = 20.188, p < .001, η2 = .419). Follow-up t-tests revealed faster reaction times when matching natural food with the self versus stranger (t(28) = −7.037, p < .001, d = 1.32), the self versus the friend (t(28) = −6.305, p < .001, d = 1.191), and when matching natural food with the friend versus the stranger (t(28) = −4.706, p < .001, d =. 889). For transformed food items, there was also a significant advantage for self-matching versus to the stranger-matching condition (t(28) = −3.497, p = .002, d =. 661). For rotten food items, results revealed faster reaction times for the self-pairing condition versus the friend-matching condition (t(28) = −5.657, p < .001, d = 1.0691), as well as an advantage for the friend versus the stranger-matching condition (t(28) = 3.843, p = .001, d =. 7263).
Importantly, these results revealed a significant advantage in reaction times when pairing the food items with the self as opposed to the friend or the stranger (Fig. 4). This demonstrates that the self-pairing manipulation was effective across all types of food items. Moreover, the differences between the self and the friend labels suggest that the advantage that we observe when associating the food items with the self are not related to familiarity. Furthermore, the lack of interaction with the factor block order suggests that it is unlikely that the pairing effects in the self-related condition are related to the presentation order of the blocks in the experimental session.
Overall, then, the results suggests that when food items were paired to the self label, relative to when they were matched to the stranger or the friend label, there was a substantial advantage for matching natural food items as indicated by greater accuracy and shorter reaction times for self-related pairing relative to the other related pairing. In addition, pairing food items with the label self led to faster responses to all food items.
Furthermore, to further control that the results were not influenced by the relatively large number of excluded participants (N = 9), we re-ran the analysis of accuracy of responses and reaction times on the full sample (N = 38). These analyses show a very similar pattern of result as that of the analysis performed on the final sample. This is, associating natural food with oneself results in greater accuracy and shorter reaction times, in comparison to associating the natural food with a friend or a stranger (see Supplemental material for details of the analysis).