PSYC 300 Lecture Notes - Lecture 4: Criterion Validity, Convergent Validity, Observational Error
Document Summary
Sources of error: random and systematic error. Reliability extent to which measure doesn"t have random error: no statistical test for reliability, will never be totally free of random error, but goal is to minimize it as much as possible, consistency. Coefficient alpha average correlation among all items in scale: range: 0. 00-1. 00 (realistically will never reach 1. 00) If nominal, use kappa: actual score = true score + random error, reliability = true score/actual score, increasing reliability. Post hoc - (item deletion) getting rid of the bad items that have low correlation. 15 items is fine, but cap it at 20. Construct validity: subjective methods face validity and content validity, objective methods. Convergent validity measures of the same conceptual variable should be positively correlated and as close to 1. 00 as possible. Discriminant validity measures of different conceptual variables should be uncorrelated, as close to 0. 00 as possible. Is the research ok: measure must have reliability in order to have validity.