Experimental Design

a) Food rating task. Participants rated all food images and their corresponding Nutri-Scores in terms of taste, health, wanting and perceived caloric content on a continuous scale b) Trial sequence of food choice task. In each trial, participants made a binary choice between two food options represented by food image and corresponding Nutri-Scores; Feedback and fixation-based fixation dots were implemented c) Experimental procedures; blue refers to sated, yellow to hungry condition (order counterbalanced). VAS refers to visual analogue scale used to assess subjective feelings of hunger. PANAS refers to a questionnaire assessing mood (see SOM1). FEV II refers to a questionnaire assessing eating behavior (see SOM2);

*indicates that these steps were only required in the first session.

Behavioral Results

a) Manipulation check: The green boxplot displays the difference (hungry-sated) in hunger state at arrival at the lab, yellow and blue boxplots display the difference (last timepointfirst timepoint) in hunger state in the hungry and sated condition, respectively. b) RT quantile plot displaying the cumulative probability of tasty (dashed lines) and healthy choices (solid lines) separately for the two conditions (quantiles are .1, .3, .5, .7, .9 of choices). c) and d) Probability to choose the left option as a function of taste and health value difference (leftright), respectively. Importantly, the dependency of choice on health information was eliminated under hunger. e) and f) Corresponding mean RTs as a function of taste and health value difference, respectively. For illustration purposes, value differences were segmented into 25 bins, and a locally weighted scatterplot smoothing technique was applied with a span of 0.75. Plots c-f) are based on all trials. Transparent shades indicate the standard errors of the smoothed choice probability and RT for the respective value bins (see also Figure S5.

Eye-Tracking Results

a) Dwell time difference between the tasty and healthy option was positively associated with the probability of choosing the tasty option in both conditions. b) The average probability to look at food image (taste attribute) compared to Nutri-Score (health attribute) was even higher in the hungry than sated condition. c) Path diagram with posterior means of the parameters, associated 95%-credible interval in squared brackets.

Quantitative Model Comparison

Posterior Predictive Checks maaDDM2ϕ

Quantile plots of simulated data with fitted parameters of the maaDDM2ϕ in blue (sated) and yellow (hungry) with HDIs of each quantile (vertical lines) and behavior. Posterior predictive checks were performed by drawing 1000 parameter values from the individual posterior parameter distribution to simulate the new data.

Parameter estimates of maaDDM2ϕ

Group parameter estimates (blue = sated, yellow = hungry; left panels) and the effect of hunger state (gray; right panels). Dashed black lines indicate the 95% HDI. a) Estimated taste weights. In both conditions the weight is larger than .5, indicating a higher weight on taste compared to health. This preference was even stronger under hunger. b-f) Parameter estimates of d, nDT, α, θ and ϕT, and the corresponding effects of hunger state. g) Parameter estimates of ϕH and the corresponding effects of hunger state, showing that the attention-driven discounting of health information was amplified under hunger.