Fig. 3: Experimental design and procedure in testing context effects under different sources of noise.

a A 2-by-2 design matrix was used to independently vary the levels of early and late noise. b Early noise was manipulated by presenting the consumer goods either physically (handled and viewed) or only as cartoon images. This induced two different degrees of precision in object valuations. c To reveal the subjective value and the noise associated with the representation of each item, participants were asked to BDM-bid on each item three times. d To build the choice sets for the subsequent task, a fixed set of the targets (V₁ and V₂) was constructed from the six highest-bid precisely presented items, resulting in \({C}_{6}^{2}=15\) combinations of target pairs. In each trial, one target pair was combined with one distracter (V₃) selected from twenty-four lower-bid items, where half of them were precisely presented items. e In the choice task, the participants were asked to choose their preferred item from a triplet under low- (10 seconds) and high- (1.5 seconds) time-pressure blocks. All images were presented as cartoons, even when they were precise items to control for visual confounds. The positions of the options were randomized across trials. f The effect of the manipulation on early noise confirmed that the standard deviation of bidding increased faster with the bid mean in the vague items than in the precise items (***, 55 participants, linear mixed regression: β = 0.049, s.e. = 0.011, t = 4.52, p < 0.001, standardized coefficient = 0.13, 95% C.I. = [0.05, 0.20]). Each dot indicates an item; thin lines indicate fitting to each individual, and thick lines indicate group averages. g The effect of the manipulation on late noise confirmed that the participants’ choice accuracy decreased when time pressure increased (***, 55 participants, within-subject ANOVA: F_{1, 162} = 39.17; p < 0.001, partial η² = 0.19, 95% C.I. = [0.11, 1.00]). Bar heights and whiskers represent the mean and s.e. of choice accuracy; the distribution is attached on the left of each bar. (Human-shaped and hourglass icons are created in BioRender⁸⁴.