The accessibility and applicability of knowledge: Predicting context effects in National Surveys

Using data from the National Health Interview Survey on Disability from 1994 and 1995, this research demonstrates that the size of accessibility effects (increased likelihood of using information activated by initial questions in responding to subsequent questions) can be modeled as a function of the applicability of the initial to the subsequent questions. When respondents reported a disability and were asked about the main condition causing the disability, they were more likely to report conditions they had been asked about earlier in the interview than alternative conditions. This accessibility effect was inversely related to the effect on reports of otheror unclassifiable conditions. The more reports of primed conditions, the fewer reports of unclassifiable conditions. A log-linear model of the accessibility bias fit the data for all disabilities. For reports of specific conditions, a measure of the applicability of context accounted for 74.4 percent of the variance of the accessibility bias; for unclassifiable or "other" conditions, it accounted for 61 percent. When limited to "well-defined" disabilities, applicability accounted for 91.9 percent of the variance (a multiple correlation of .96). Finally, models of the context effects derived from the 1994 data were tested against the actual effects for the 1995 data. The correlation between predicted and actual effects was .80 across disabilities. The theoretical and the practical implications of the findings are discussed.