Study 1 & 2
Study 1 aimed to compare the feasibility and effectiveness to increase healthy food choices when composing a colourful meal to a typical, a healthy and a low calorie meal. Study 2 aimed to further distinguish the ease and consequences of promoting a colourful meal versus a varied meal. The typical meal was again included to provide a common comparison condition between studies. Both studies were conducted in a controlled experimental setting, using a realistic Fake Food Buffet [39,40,41,42].
Samples
For Study 1, a power analysis using G*Power 3.1 [48] to detect a small to medium effect (Cohen’s f = .15) in a within-subjects design with four measurements yielded an N of 62 for 80% power. Eighty-four participants were recruited through the university online study pool. Everyone in the pool was eligible for participation unless they had defective colour vision or had taken part in previous studies with Fake Food buffets. One participant had to be excluded because of a slight impairment of colour vision, reducing the final sample to N = 83 (83% female). The sample had a mean (M) age of 22.11 (standard deviation (SD) = 2.89) and a mean body-mass index (BMI) of 22.15 (SD = 3.27, range 16.81–38.77). All participants except one were students representing a range of academic majors including Psychology (64.6%), Politics (8.5%), Linguistics, and Teacher Training Programs (4.9% each). Other academic majors were represented by less than 4% of the sample.
For Study 2, a power analysis using G*Power 3.1 [48] to detect a large effect (Cohen’s f = .4; c.f. results Study 1) in a within-design with three measurements yielded an N of 12 for 80% power. Forty-two participants (76% female) were recruited using the same procedure as Study 1. The sample had a mean age of 22.21 (SD = 6.24) and a mean BMI of 21.54 (SD = 2.67, range 17.06–30.19). All participants except one were students, with the majority studying Psychology (81%).
In both studies, participants received 1 h of course credit or 10€ as compensation.
Design and procedure
The studies were approved by the University of Konstanz ethics committee and carried out in accordance with the Declaration of Helsinki and the guidelines of the German Psychological Society. Participants were invited to the laboratory for individual sessions and gave written informed consent. Both studies followed a within-subjects design, where participants were initially provided with tableware and asked to serve themselves a meal that they would typically have for lunch from a Fake Food Buffet. When they were finished, they were asked to place the dishes on a serving tray and fill in a short questionnaire. In Study 1, participants were then asked to serve themselves a healthy, a low calorie, and a colourful meal in random order. The buffet was restocked after the second meal. In Study 2, the buffet was restocked immediately after participants self-served the typical meal, and they were then instructed to serve themselves a varied meal and a colourful meal, in random order. Finally, in both studies, participants filled in a questionnaire assessing demographics and evaluations of the choice strategies, while the experimenter unobtrusively weighed and counted the Fake Food items. The participants were then debriefed and paid.
Materials and measures
All items used in this study are listed in Additional file 1.
Fake food buffet and food choice
The Fake Food Buffet was derived from Sproesser et al. [42] (see also Bucher et al. [39], Bucher et al. [40], Mötteli et al. [44] for similar buffets), with the addition of vegan falafel and tofu sausages. The buffet included a total of 74 different food items which were placed in serving bowls and arranged on a table to resemble an actual buffet (see Fig. 1). Participants were given a serving tray (55 cm × 35 cm) with a large and a small plate (27 and 21 cm in diameter respectively) and a small bowl (12 cm diameter). The components of the self-served meals were weighed (continuous items, e.g. peas) or counted (e.g. strawberries). The amount of food replicas was converted into the respective amount of real food by multiplying the amount of each replica with a predetermined factor based on a comparison of the replica item and the respective real item (see Sproesser et al. [42]). The foods were grouped into eight categories (vegetables, fruits, grains and starches, protein sources, dairy, fats, sweet extras, and drinks), and standardised to the total weight of the meal according to König and Renner [38].
Fig. 1
Fake Food Buffet used in Studies 1 and 2
Manipulation check
After each meal, participants were asked to indicate whether they chose foods that were colourful. In Study 1, participants were additionally asked to indicate whether they chose foods that were healthy or low in calories, while in Study 2 they were additionally asked to indicate whether they chose foods that were varied. All items used a six-point Likert scale ranging from (1) “I do not agree at all” to (6) “I totally agree”.
Evaluation of the choice strategies
After each meal, participants were asked to indicate how filling the self-served meal would be on a six-point semantic differential from (1) “not at all filling” to (6) “very filling”. After the participants had chosen all meals, they rated the strategies’ feasibility ((1) “very difficult” to (6) “very easy”) and simplicity ((1) “very complex” to (6) “very simple”) and indicated if eating in accordance with the strategy was fun ((1) “not at all fun” to (6) “very fun”) on six-point Likert scales. Furthermore, participants were also asked to rank the choice strategies according to their anticipated taste and feasibility in daily life. A ranking task was used to avoid ceiling effects, as it could be expected that participants generally self-serve tasty meals.
Statistical analysis
Data was analysed using IBM SPSS (Version 25). In Study 1, missing values were 1.2% for the evaluation of the healthy and colourful meals due to missing questionnaires and 1.2% for the rankings. Within-subjects Analyses of Variance (ANOVAs) were computed to compare the strategies regarding the overall size, proportion of food groups, and evaluation by the participants. Significant results were followed up by Bonferroni paired comparisons. For all tests, α was set to .05.
Study 3
This study aimed to implement and test eating colourful meals to facilitate healthy food choices using a smartphone-based Ecological Momentary Intervention.
Sample
Sample size estimation in intensive longitudinal studies is difficult when little information about the effects of interest is available [49], so N = 108 participants were recruited in accordance with a previous study [38]. Three waves of participants were recruited using an online study pool with each wave containing n = 46, n = 34, and n = 28 participants, respectively. All subjects were eligible for participation unless they had defective colour vision, or had taken part in previous studies assessing perceived meal colour variety. Several participants had to be excluded (1) because they did not complete the study (n = 4), (2) because they had difficulties using the study app (n = 1), (3) due to data loss because of incorrect settings on the smartphone (n = 2), or (4) due to data loss from a server error in the second recruitment wave (n = 21).
The final study sample consisted of N = 80 participants (88% female) aged from 18 to 43 years (M = 22.41, SD = 4.00). Their mean BMI was in a normal range (M = 22.86, SD = 3.52, range 18.04–37.47). There were no differences in age, gender, or BMI across recruitment waves (age: F(2,77) = 0.99, p = .377; gender: χ2(df = 2) = 3.40, p = .183; BMI: F(2,76) = 0.81, p = .449). Ninety-nine percent of participants were students: Psychology (51%), Teacher Training Programs with various majors (8%), Law (5%). Other academic majors were represented by less than 5% of the sample. Participants received 2 h of course credit or 20€ as compensation.
In total, N = 1,327 meals were logged, but recorded data were incomplete for n = 117 meals (e.g. due to missing pictures). Therefore, the present analyses were conducted on N = 1,210 unique meals.
Design and procedure
The study was carried out in accordance with the Declaration of Helsinki and the guidelines of the German Psychological Society and was approved by the University of Konstanz ethics committee. The study used a single-group within-subjects design. Lunch meals recorded during the first week represent the baseline food consumption. During the second week of the study (intervention period), participants also received a daily prompt reminding them to eat a colourful lunch (“Eat a colourful lunch meal today.”). The time they received the prompt was tailored to the individual by sending it to each participant at the time they stated that they usually bought or prepared their lunch.Footnote 1 During the third week (follow-up), participants again recorded their lunches but without receiving any prompts.
Prior to the study period, participants were invited to the laboratory for individual sessions. They were informed about the study procedure and gave written informed consent. Participants with Android smartphones (n = 38) were then asked to install the smartphone application (app) movisensXS (movisens GmbH Karlsruhe; version 0.8.4203; available on Google Play) and download the questionnaires, while participants without an Android smartphone (n = 42) received a smartphone (ASUS Padfone Infinity or Motorola Moto G 1st generation) with the app and questionnaires installed. Furthermore, height and weight were measured. The first time they used the app, participants completed a pre-study questionnaire assessing demographic variables and indicated the time of day they usually prepared or went to have their lunch.
The participants were then asked to record their lunch meals in real life for 3 weeks starting the following day by (1) taking a picture (see Fig. 2 for examples), (2) describing the meal, (3) rating the meal’s colours, and (4) taking a picture of any leftovers. Additionally, participants were able to record missing events by indicating (1) that they forgot to record their lunch or (2) that they did not have lunch that day by pressing the relevant button on the app’s home screen (Ziesemer K, König LM, Boushey CJ, Villinger K, Wahl DR, Butscher S, Müller J, Reiterer H, Schupp HT & Renner B: Occurance of and reasons for “missing events” in a mobile dietary assessment: results from three event-based EMA studies. Submitted). Questionnaire data and food pictures were transferred to the server by mobile data or Wi-Fi connections.
Fig. 2
Examples of meal pictures taken by participants in Study 3
After 3 weeks, participants were asked to fill out a post-study questionnaire to evaluate the ease and enjoyment of the prompt. Subsequently, they returned to the laboratory where their weight was measured again, and they were compensated for participating.
Materials
All items are listed in Additional file 1.
Perceived meal colour variety
Participants rated the meal’s colour on a 100-point visual analogue scale ranging from ‘one colour’ to ‘many colours’ (see also König and Renner [38]).
Food intake
Food intake was coded by trained research staff using the participant provided meal descriptions and food pictures following a previously developed coding manual [38] that is based on German dietary guidelines [50]. All foods were assigned to one of seven food groups (vegetables, fruit, grains and starches, animal and other protein sources (i.e. ‘protein’), dairy, fried foods, and desserts and other sugary foods (i.e. ‘sugary extras’)) and their serving sizes were determined based on the pictures taken before and after the meal. As in König and Renner [38], a final food intake score was computed by dividing the serving sizes of all seven categories by the total amount of portions per meal, representing the proportion of the given category in the whole meal.
Evaluation of the prompt
After the 3 week study period, the prompt to eat a colourful lunch was evaluated on two 100-point visual analogue scales. Participants indicated whether they found it easy (“Eating colourful meals was easy.”, (0) “I do not at all agree,” (100) “I fully agree”)/ pleasant (“Eating colourful meals is pleasant.”, (0) “I do not at all agree”, (100) “I fully agree”) to eat colourful meals. Participants were also asked to indicate whether they paid attention to the prompts on a 100-point visual analogue scale to assess perceived compliance (“I paid attention to the prompts that I received during the study.” (0) “I do not at all agree”, (100) “I fully agree”).
Demographic variables and BMI
When using the app for the first time, participants were asked to indicate their gender, age, current occupation, field of study, and dietary habits. BMI was calculated from measured height and weight. Participants wore light indoor clothing and were asked to remove their shoes before being weighed. Height was measured before the study using a wall-mounted stadiometer, and weight was measured before and after the study using a digital scale (Omron Body Composition Monitor, BF511).
Statistical analysis
Following the procedure previously described in König and Renner [38], data was analysed using multilevel linear modelling [51] in R 3.2.3 with the packages lme4 version 1.1–11 [52] and lmerTest 2.0–30 [53]. For all analyses, individual meals defined Level 1, which were nested within participants (Level 2). To analyse the relationships between perceived meal colour variety and intake of the seven food groups, perceived meal colour variety was entered as a Level 1 predictor and thus group-mean centred [54]. Differences in food consumption between baseline, intervention, and follow-up weeks were analysed as a function of time. Models were computed separately to evaluate the difference between baseline and intervention weeks and the difference between baseline and follow-up weeks. Following the procedures suggested by Lischetzke et al. [55], time was dichotomized into (0) baseline and (1) intervention week, and (0) intervention and (1) follow-up week, respectively.
For all analyses, both random slopes and random intercept models were then computed and compared using a deviance test [51]. If the deviance test was significant, differences between participants in the strength and/ or direction of the relationship were assumed and the percentage of positive and negative slopes was computed [51]. For all multilevel models, quasi-R2 was calculated as an estimate for the effect size, comparing the preferred model to the intercept only model.
Choice strategies were compared using within-subjects ANOVAs with Bonferroni-corrected post-hoc comparisons.
