Reliability problems of the datum: Solutions for questionnaire responses
ERIC Clearinghouse on Assessment and Evaluation
EDRS No. ED451238. Paper presented at the European Meeting of the Psychometric Society (11th, Luneburg, Germany, July 1999).
Questionnaires often ask for estimates, and these estimates are given with different reliabilities. It is difficult to know the different reliabilities of single estimates and to take these into account in subsequent analyses. This paper contains a practical example to show that not taking the reliability of different responses into account can lead to erroneous conclusions. A solution is suggested in which two estimates are requested that are then used as upper and lower bounds. The mean of this double estimate then acts as equivalent to, or more accurate than, the traditional single response, and the range can be used to calculate 'more credible measures.' A particular example of a more credible measure is the mid-double estimate (mean of the two estimates) corrected for its unreliability by dividing it by the difference in these two estimates. Other more credible measures, based on double estimates, are suggested for regression and correlation analyses.....