We develop and test an economic model of the cognitive and non-cognitive foundations of survey item- response behavior. We show that a summary measure of response behaviour – the survey item-response rate (SIRR) – varies with cognitive and less so with non-cognitive abilities, has a strong individual fixed component and is predictive of economic outcomes because of its relationship with ability. We demonstrate the usefulness of SIRR, although an imperfect proxy for cognitive ability, to reduce omitted-variable biases in estimated wage returns. We derive both necessary and sufficient conditions under which the use of an imperfect proxy reduces such biases, providing a general guideline for researchers.