1 Bring in the responses
One row per learner, one column per item, values 0 or 1. Commas, tabs or spaces all work. Leave a cell empty for an item that was not administered.
2 Calibrate the items
Item parameters have to be estimated before anyone can be scored. Two parameters is the safer default; add a guessing parameter only if the items are multiple choice and you have enough low scorers to identify it — step 3 will tell you whether you do.
| ITEM | PROPORTION CORRECT | α SLOPE | β LOCATION | γ AT θ = 0 | γ UNDER A FLAT PRIOR |
|---|
Information rises without limit as θ approaches 1, because the model fixes P(1) = 1 there. That is a property of the construction, not precision about the ablest learners — which is why the scores in step 4 carry a posterior standard deviation rather than an information-based standard error.
3 Read the warnings
These are the conditions under which an estimate should not be reported as it stands. They are checked automatically because they are easy to miss and expensive to get wrong.
4 Score the learners
Each row shows the estimate as a mark on the mastery rule, with the band covering one standard deviation either side. Read the band before the number.
| # | RAW | MASTERY | ESTIMATE | SD | NOTE |
|---|