Select by clicking on "Test Randomness" or from the Graphs menu. Empirical intervals are set with the "Empirical Interval" slider. This displays the local value of the mean-square statistics. The Outfit mean-square statistic (standardized residual chi-square divided by its degrees of freedom) is the red line. The Infit mean-square statistic (ratio of observed to expected residual variance) is the blue line.
The Test Empirical Randomness shows the average mean-square for the subset of responses in a (person ability - item difficulty) interval on the latent variable. For instance,
Are the noisy responses by people with high abilities relative to the item difficulties (= carelessness) = high mean-squares for high measures?
Are the noisy responses by people with low abilities relative to the item difficulties (guessing) = high mean-squares for low measures?
Are the responses too predictable for people with abilities far from the item difficulties (= Binet-style response imputation) = low mean-squares for high measures and for low measures?
In classical test theory, 3-PL IRT, and conventional statistics, low mean-squares are considered good. In Rasch theory, they indicate some redundancy in the responses, but they do no harm.
Buttons are described in Graph window.
