Study design
An online survey was developed to investigate the portion size norm of commonly consumed discretionary foods in Australia. Online image-series for 15 discretionary foods were validated against corresponding real foods. Using a cross-sectional, within-person crossover design, participants reported their normal portion sizes for each test food twice, once based on food images and the other time based on real foods.
Selection of test foods and portion sizes
A variety of commonly consumed discretionary foods, familiar to Australian consumers (23), were included; sweet snacks (M&Ms, chocolate bars, chocolate blocks, and sweet biscuits), cakes (layered cake, caramel slices, muffins, and banana bread), savoury snacks (savoury biscuits and potato crisps), fast foods (pizza, nuggets, and hot chips), and sugary carbonated drinks (in glasses or cups, and in bottles or cans).
Eight portion sizes in increasing size were included for each food (except for drinks in bottles or cans where six options were included). This is based on literature that suggests presenting a range of serving size options may assist with portion size estimations (21) and an even number of options helps to avoid the temptation of choosing the centre image (21, 24). Detailed criteria used to develop the portion size options and the portion size weights are available in supplementary materials (Appendix 1).
Questionnaire design
The accompanying questionnaire was developed using Qualtrics (an online survey development software) and consisted of three sections: demographics, food images, and real foods. The demographic section collected information on participants’ gender, age, self-reported height and body weight, postcode of home address, usual physical activity level (PAL), and confidence in their cooking skills. PAL was estimated using the physical activity factor and classified as sedentary, lightly active, moderately active, very active and extra/vigorously active (25, 26). Confidence of cooking skills was assessed as a marker of food literacy/awareness of food quantity (27) using a validated Likert scale (28).
The food image section displayed the eight successive images corresponding to the sliding scale question, labelled from smallest ‘1’ to largest ‘8’ and additional selections of ‘0 – I do not eat this food’ and 9 – greater than the largest option displayed’ (Fig. 1a). Participants were instructed to move the marker to their corresponding portion size norm, which would become enlarged for easier viewing. A cover photo of a typical manufactured package was used as an example to orient the participant to each food. The JavaScript code was based on Embling et al.’s image carousel (22).
The real food section consisted of the questions and sliding scale identical to the image section (Fig. 1b) but without the food images. Participants answered each question by observing the labelled portion size options present at the food stations.
The presentation order of section and test foods within sections were both randomised using a built-in randomiser. The survey questionnaire was pilot tested in the target population (March 2022) and minor modifications were made to improve usability. Further details of study and questionnaire design are attached as supplementary material (Appendix 1).
Study procedure and participant recruitment
A convenience sample of university staff and students was recruited through online advertisements and the distribution of physical flyers. An online screener questionnaire excluded participants who did not meet the following criteria: living in Australia, aged between 18–65 years, fluent in English, no current or previous diagnosis of an eating disorder, and who were able to attend an in-person laboratory session. Participants attended the in-person session at a university campus in Sydney (April to May 2022) and were instructed to complete the questionnaire individually using a laptop. Participants were reminded of the definition of portion size as ‘the amount of food they eat at one sitting’. Researchers remained in the laboratory room in an unobstructive manner during the study process.
A small token was offered to compensate participants’ time. The study was approved by The University of Sydney Human Research Ethics Committee (ethics approval number 2022/147). Study protocol was registered in priori on the Open Science Framework (OSF registration DOI: osf.io/x3fm7).
Due to the preliminary nature of this study, a power analysis could not be calculated based on previous literature. Thus a sample size of 100 participants was used as recommended for preliminary validation in dietary assessment (29).
Statistical analysis
Data were analysed using IBM SPSS v28 (IBM, Armonk, NY, USA). Descriptive analyses on participants characteristics was conducted. For each food, data were excluded if participants reported that they do not consume the food. Reported normal portion sizes from two methods were compared using cross-classification and intra-class correlation coefficients (ICC, two-way mixed model, average measure). Data were classified as correct match (described as participants who selected the same image option as real foods), adjacent match (described as participants who selected a portion size image one option away from what they selected based on real foods), and gross mismatch (described as participants who selected a portion size image four or more options away from what they selected based on real foods) (30). ICC values < 0.5 were considered poor, 0.5–0.75 moderate, 0.75–0.9 good, and > 0.9 excellent (31). Proportion of over- and underestimation was tested based on real foods being the reference standard. The relationship between cooking confidence and the ability to match images with real foods (that is, the mean proportion of correct match across foods, per participant) was investigated using the non-parametric Chi-square test. The median energy (in kilojoules) of reported portion sizes were calculated based on the nutrition information panel on food packages.