The seven effects, in plain words
What the study measures
Seven choice-set context effects: adding or relabeling an option in a multi-attribute choice set shifts the shares among the options that were already there. Each is paired with one registered statistic, computed against a matched control cell, and each must first meet its own human replication criterion: an effect that does not replicate in our calibrated sample exits fidelity scoring and is reported as a replication result in its own right.
A registered counterbalance cuts across the frequency, range, and phantom structures: the decoy targets the premium (high-desirability) option in five categories and the budget (high-feasibility) option in the other five, with a registered directional prediction that the premium-target effect is the stronger one. Six of the seven tests share one binary control cell per category and tier, so all confirmatory tests are Holm-corrected within families of effects.
The registered stimuli, as participants saw them
Stimulus explorer
Pick a product category, then a choice structure, and see the exact choice set. The same single stimulus file feeds the surveys, the model calls, and the analysis, so what you see here is what humans and models see. Every value is final: the three-wave calibration pre-test is complete.
The Numbers · Words · Pictures toggle shows the same stimulus in its three formats where the format module runs. Only the presentation changes, so a model that shows an effect on the numbers but misses it in a picture is reading text arguments where a person would perceive the options.
Hover an option's small role label for what it does in the set. The format module (words and pictures) runs in four categories (Bluetooth speakers, hard drives, televisions, water bottles) for the baseline pair, the frequency-decoy set, and the compromise set. The pictures format runs in three of the four: the hard-drives pictures cells were removed at calibration, where that pair stayed beyond the registered limits in all three waves; the removal is reported on the calibration page.