Jump to start of content

Breadcrumb trail:

An investigation of the performance of test items developed through two item modeling methods.


Grosso LJ, Dugosh JW, Zhang Y, Lelieveld C. — American Board of Internal Medicine

Presented: American Educational Research Association Annual Meeting, April 2013

Abstract: Testing programs need reasonably sized item pools, well balanced with regard to the content and statistical specifications defined in the test plans. Modeling is an item engineering method to efficiently and cost-effectively increase item development. Two item modeling methods were studied (Method 1 – changed the item stem so that a different answer from that of the source item was correct; Method 2 – changed the item so that it poses the same task and keeps the same answer, but no longer “looks” like the source item). Results revealed that the Method 1 pretest item survival rate was similar to that of the new items, while the Method 2 survival rate was better. Results also showed that although statistical performance for source and models can differ markedly, Method 2 items tended to perform more like their source than the Method 1 items. Findings suggest that item modeling provides an efficient way to generate new test items and effectively fill content area shortages.

For more information about this presentation, please contact Research@abim.org.