The relative contribution of response bias and weighting-of-similarity bias to valence asymmetry in attitude generalization

Negative generalization asymmetry (NGA) is a tendency to generalize negative attitudes more widely than positive attitudes. Studies found robust NGA for new objects that resemble both positive and negative learned objects, and even stronger NGA for new objects that resemble neither. Two biases were suggested to underlie NGA: (1) negative response bias, whereby a novel object is perceived as novel, but forced to make a dichotomous good/bad decision, a responder has a strategy to classify it as bad; (2) negative weighting-of-similarity bias, whereby a novel object seems more similar to negative learned objects than to positive learned objects. We hypothesized that adding a third response alternative, “cannot be evaluated”, to the good/bad response alternatives commonly used in the BeanFest paradigm would reduce negative response bias but not negative weighting-of-similarity. As predicted, the new response alternative was used more as similarity to learned stimuli decreased. Also as predicted, NGA, although still significant, was lower in the 3-alternatives condition. Interestingly, extent of reduction in NGA due to adding the new response alternative did not vary as a function of stimulus type, and was not particularly strong for stimuli very dissimilar to any learned stimuli. These results confirm that NGA is indeed partly due to response bias. They additionally suggest that increased NGA with decreasing similarity to learned stimuli cannot be attributed solely to negative response bias. Thus, avoidance of new, unfamiliar objects does not reflect only strategic caution, but rather reflects to a large extent a naïve-realistic belief that they are negative.