Crowding distorts features, not objects: structured target confusions in a 4AFC conjunction task

Visual crowding is the disruption of peripheral object identification by nearby flanking clutter. It remains unclear whether crowding disrupts the bound representation of an object or degrades the object's component features separately. We addressed this with a 4AFC conjunction-report task in which observers reported both the orientation and the brightness of every target. Target Gabors at 6 degrees eccentricity were presented alone or flanked at 1, 3, or 5 degrees separation by Gabors varying independently in colour and/or orientation similarity to the target. We analysed the data three ways: joint accuracy, accuracy on each feature alone, and the statistical coupling between the two report channels. Orientation reports were principally impaired by orientation-similar flankers within the smallest target-flanker separations, whereas luminance reports were impaired by target-colour-similar flankers across a broader spatial range. Observed joint accuracy was closely predicted by the product of the two marginal accuracies in every condition. A complementary chi-square analysis of the full 4x4 stimulus-response matrices agreed with this conclusion at the most-crowded conditions, while revealing structured response biases that the ratio test could not detect. These biases were present at baseline and were not amplified by crowding. The conjunction errors observed under crowding therefore arose from two largely independent feature-specific failures, rather than from disruption of a representation in which the two features are bound. These results indicate that object-level crowding errors in this paradigm are the downstream signature of two independent feature-level processes rather than the breakdown of an object-level mechanism.