Jon Krosnick - Work in Attitude Research

Work in Attitude Research

Krosnick has investigated in detail how attitudes, in general, are formed and how they relate to responses to surveys. He has modeled the emotional aspect, affect, that influences attitudes in a framework for long-term memory drawn on the computer model of short-term random access memory and longer-term disk storage. Long-term memory is posited to be made of interconnected nodes, and Krosnick models affect as tags attaching to the node for say a political candidate, weighting it and influencing other nodes through connections. The well-informed and politically savvy are expected to have more well-developed network structures of such nodes.

Krosnick has also researched attitude strength, which per him is a subjective element, with one possible measure being the attachment to a topic a respondent expresses in a self-report survey. He showed this form of attitude strength has four disparate dimensions, revealed by the statistical technique of factor analysis. The four dimensions found were polarized and positive or negative intensity (valence) of attitudes, ease of retrieval of the associated memories (accessibility), personal beliefs driving attitudes, and degree of thinking done on the subject.

On the practical issue of how attitudes affect survey results, in line with other studies, Krosnick has looked separately at well-informed subjects aware of political issues and ill-informed or unmotivated respondents. In his research jointly with colleagues he found knowledgeable subjects used different cognitive organized patterns of thought (schemas) and knowledge-churning strategies from the naïve or undermotivated subjects. Non-intuitively, in certain circumstances the experts were easier to prime with specific appeals or political advertisements. The other group tended to generate more evasive answers avoiding the question, especially when the issue was not considered relevant. Some biases arising from this were a tendency to settle on the midpoint of a scale with an odd number of divisions, being more influenced by leading questions, and answering most questions with the same number on a scale, especially toward the end of the survey, a form of satisficing. These combined increased the chance and amount of measurement error for such responders.