If a measure cannot be consistently reproduced from two assessments, it cannot accurately measure an underlying construct. 80, even if the concepts they measure were perfectly correlated (1.00), would be only. The maximum possible correlation between two scales with coefficient alphas of. Thus, if r R is the reliability coefficient for a measure of a variable and r V is the reliability of a criterion variable, the correlation (r RV) between the measure and the criterion has the following limit: /2r RV/2 < (r V)* (r V) /.Įmpirically, the maximum correlation between two scales, used to measure the criterion or construct validity of one of them, is defined as the theoretically maximum correlation multiplied (reduced) by the square root of the product of the respective scale’s reliability coefficients (test-retest or coefficient alpha). Reliability sets an upper bound to validity. Formulas exist to measure dimensions of reliability. The procedures for evaluating an instrument used at two times are straightforward. If an instrument does not measure what it should be measuring, it is irrelevant that the measurement is reproducible. Although developers and users of instruments must be concerned about both validity and reliability, validity is more important than reliability.
