To determine the best-fit model, which comparison should be minimized?

• A. predicted true scores and observed scores
• B. observed scores across assessments
• C. predicted error scores and observed scores
• D. true scores and error scores

Answer: (A)

In true-score thinking, the goal of a best-fit model is to explain as much of what is observed as possible by estimating each person's true score. The observed score equals the true score plus error, so a good model should produce predicted true scores that line up with the actual observed scores. When the predicted true scores closely match what we observe, the remaining difference—essentially the error component—is minimized, signaling a strong fit.

That’s why matching predicted true scores to observed scores is the best criterion for model fit. The other options don’t assess how well the model captures the underlying true ability: looking at scores across assessments examines stability rather than fit; predicting and comparing error scores isn’t the standard way to gauge how well the latent construct is being captured; and comparing true scores to error scores mixes components that are supposed to combine to form the observed score rather than evaluating fit.