Cognitive and non-cognitive skills are important determinants of labor market outcomes, but are often unobserved. We propose a proxy for these skills derived from item non-response information and a procedure to test its validity. Exploiting a unique data-collection feature of an Australian survey, we find that fraction answered on a self-completion questionnaire fulfils all necessary requirements to be a valid proxy for cognitive skills and outperforms alternative proxies derived from paradata. Fraction answered captures a third of the effect of cognitive ability on wages and education. We provide a simple solution to reduce omitted-variable biases by up to 11%.