Construct Generalization is a method for investigating the fit of alternative construct theories to item difficulties obtained from tests purporting to measure the same or a similar construct. A good construct generalization study demonstrates how a number of different operationalizations of a construct (i.e., item calibrations) can be integrated and understood within a common theoretical perspective. During this process, the construct, the construct theories, and the test selection criteria are all improved.
For each construct of interest, we ask the simple question: "Is there a single equation, generated from a construct theory, that can account for the variation among item difficulties taken from a diverse set of instruments purportedly measuring this construct?"
The method of construct generalization consists of seven steps: (1) assemble a sample of relevant instruments, (2) estimate Rasch item difficulties separately for each instrument, (3) conceptualize alternative construct theories that explain variation in item difficulties and express these theories as specification equations, e.g., "reading difficulty is proportional to sentence length," (4) compute correlations separately for each instrument between the item difficulties as observed and those computed from the competing specification equations, (5) select a provisionally "best" construct theory and associated specification equation, (6) test the causal status of the variables in the specification equation by verifying that the specification equation predicts the observed calibrations for as yet untested items, (7) re-cycle steps 3-6 until satisfied with both the data-theory fit and the causal status of the variables in the specification equation.
A. Jackson Stenner
Construct generalization. Stenner AJ. Rasch Measurement Transactions 1994 7:4 p.323
Construct generalization. Stenner AJ. … Rasch Measurement Transactions, 1994, 7:4 p.323
The URL of this page is www.rasch.org/rmt/rmt74e.htm
Website: www.rasch.org/rmt/contents.htm