Development and validation of an online tool to assess perceived portion size norms of discretionary foods

The study addresses whether perceived portion size norms—individuals’ typical perception of how much of a given food to eat at one occasion—have shifted upwards due to ubiquitous large servings. Such shifts are concerning for discretionary foods high in saturated fat, added sugars, salt, and/or alcohol, which contribute to excess energy intake and chronic disease risk. Prior work distinguishes social versus personal portion size norms and proposes a norm range model where offering portions at the lower end of the perceived normal range can nudge lower intake. However, the range of perceived norms for common discretionary foods is not well established, and existing assessment methods lack standardization and validation. The aim was to develop and validate an online image-based tool to measure perceived portion size norms for common discretionary foods among Australian adults.

Multiple tasks have been used to assess perceived portion size norms, including self-selected portion size tasks with provided options, computer-based normality judgments, and estimating number of portions in packages. Existing tools often use real foods with a single option, which can induce unit and social desirability biases. Image-based tools can offer broader ranges but are seldom piloted or validated in the target population. Prior literature suggests presenting a range of options and avoiding a central default may improve estimation. There is heterogeneity in methods, foods tested, and limited evidence on factors influencing over- or underestimation. Cooking skills have been linked to better diet quality and may influence portion estimation, but this has been underexplored.

Design: Randomized within-person crossover validation conducted in a laboratory session (April–May 2022). Participants completed the tool twice for each food: (1) selecting perceived portion size norm from image-series on a computer; (2) selecting from equivalent real, weighed food options displayed at laboratory food stations. Order of methods and food presentation was randomized (Qualtrics randomizer). Participants were blinded to the study aim and reminded that portion size is the amount eaten at one sitting. Ethics approval: University of Sydney HREC (2022/147); OSF preregistration (https://doi.org/10.17605/OSF.IO/X3FM7). Recruitment and sample: Convenience sample of university staff and students in Sydney; inclusion: living in Australia, 18–65 years, English fluent, no current/past eating disorder, able to attend lab. Sample size target 100 for preliminary validation; 114 participated. Tool development: Fifteen discretionary foods selected based on national nutrition survey and availability: sweet snacks (M&Ms, chocolate bar, chocolate block, sweet biscuits), cakes (layered cake, caramel slice, muffin, banana bread), savory snacks (savoury biscuits, crisps), fast foods (pizza, nuggets, hot chips), sugary carbonated drinks (cola in cup/glass; cola in bottle/can). For each food, eight increasing portion size options (six for bottle/can drinks) were defined. Portion ranges and increments were derived from supermarket/fast-food package sizes, median and percentile typical portions from literature, and pilot feedback. One of images 3–5 approximated the survey-based median portion size. Reference objects (e.g., original package or credit-card-sized marker) were included. Tool implementation: Built in Qualtrics with an eight-image carousel (JavaScript from Embling et al.) paired to a sliding scale labeled 1 (smallest) to 8 (largest), plus 0 (“I do not eat this food”) and 9 (“greater than the largest option”). The real food section used identical questions and scales without images; participants observed labeled, weighed options at stations. Demographics: Collected gender, age, self-reported height/weight, postcode, usual physical activity level (PAL; sedentary, lightly/moderately active, very/extremely active), and cooking confidence (validated Likert scale). Statistical analysis: Analyses in SPSS v28. For each food, data excluded if a participant reported not consuming that food. Agreement between images and real foods assessed via cross-classification (exact match; adjacent match = ±1 option; gross mismatch = ≥4 options apart) and ICC (two-way mixed-effects, absolute agreement, average measures). ICC interpretation: <0.5 poor; 0.5–0.75 moderate; 0.75–0.9 good; >0.9 excellent. Over-/underestimation tested using real foods as reference. Association between cooking confidence and mean percentage of correct matches per participant assessed by Chi-square.

- Participants: 114 completed; 82.5% female; mean age 24.8 years; 77.2% within normal weight range. - Comparisons: 65–111 respondents per food; total 1442 image vs real food comparisons. - Agreement: Overall, 91% were exact or adjacent matches; 53% exact matches; <1% gross mismatches (>4 categories apart). Overall ICC across all foods = 0.85 (95% CI 0.83–0.87), indicating good agreement. - Food-specific ICCs and accuracy (examples): • Pizza: ICC 0.93 (0.90–0.96); 57% exact; 93% exact+adjacent. • Chocolate blocks: ICC 0.92 (0.88–0.94); 60% exact; 96% exact+adjacent. • Cola bottle/can: ICC 0.90 (0.84–0.94); 69% exact; 88% exact+adjacent. • M&Ms: ICC 0.87 (0.80–0.92); 51% exact; 94% exact+adjacent. • Nuggets: ICC 0.87 (0.77–0.92); 61% exact; 97% exact+adjacent. • Crisps: ICC 0.80 (0.71–0.86); 44% exact; 86% exact+adjacent; 2% gross mismatch. • Hot chips: ICC 0.79 (0.70–0.86); 39% exact; 89% exact+adjacent. • Cola cup/glass: ICC 0.77 (0.53–0.88); 43% exact; 93% exact+adjacent; 3% gross mismatch. • Moderate ICCs observed for chocolate bar 0.71 (0.51–0.82), muffin 0.72 (0.46–0.84), banana bread 0.69 (0.48–0.80). - Energy content of perceived norms (median, images vs real foods): Ranged from ~405 kJ (cola cup/glass) to ~2579 kJ (pizza). Medians similar between methods for 13/15 foods; images tended to yield higher selected portions for banana bread, nuggets, and cola cup/glass. - Cooking confidence: Higher cooking confidence associated with higher overall exact match rate (46% vs 57%; p=0.04). Significant differences for chocolate blocks (p=0.05), chocolate bar (p=0.03), crisps (p=0.02), hot chips (p=0.02). No significant order effects on agreement.

The online image-series tool demonstrated high agreement with equivalent real food options, addressing the need for a validated, scalable method to assess perceived portion size norms for discretionary foods. The high rates of exact or adjacent matches and good overall ICC suggest that image-based assessments can capture individuals’ perceived normal portions reliably across a broad range of discretionary foods. Variation in agreement across foods likely reflects complexities in portion judgment, including heterogeneity of serving sizes in the food environment and visual challenges in interpreting two-dimensional images for certain items (e.g., cakes, chocolate bars). The positive association between cooking confidence and agreement suggests that food literacy may enhance portion estimation accuracy, aligning with literature linking cooking skills to better diet-related outcomes. The tool’s design choices—broad option ranges, piloted increments based on real-world packages, randomized presentation, and inclusion of reference objects—likely mitigated social desirability and unit biases. Findings support the use of the tool to characterize current portion size norms, which can inform interventions aiming to nudge selections toward the lower end of perceived normal ranges to reduce overconsumption.

An online, image-based tool assessing perceived portion size norms for 15 discretionary foods showed good agreement with real food options and is suitable for research on portion norms across eating contexts. Understanding current norms can inform public health messaging, food labeling, and industry practices to align offered serving sizes with consumer norms and gradually nudge toward smaller portions. Future work should validate and extend the tool to other populations, explore moderators such as age, gender, and socioeconomic status, and expand to healthier food groups (e.g., vegetables, grains).

- Presentation context: Foods displayed on white plates may not reflect typical real-life consumption contexts. - Single-session crossover: Completing both methods in one session may introduce carryover or memory effects, though order was randomized and no order effect was detected; participants were unaware options were identical across methods. - Sample representativeness: Convenience sample of primarily young, female, highly educated university-based participants with higher-than-average proportion in normal weight range limits generalizability. - Image constraints: Two-dimensional, non–life-sized images may reduce accuracy for certain foods (e.g., cakes, bread-like items, chocolate bars), potentially compounded by small weight increments between options. - Food environment variability: Large variability of serving sizes across retail settings may complicate generalization of perceived norms. - Some foods showed only moderate ICCs, indicating room for refinement of image-series for those items.