The Item Statistics section of the Aware Local Test Analytics tool is perhaps the most informative section for discussions about how well a test performed. This section contains a number of metrics for each individual item that can help identify potential error in the test design itself.
Additionally, this section provides clues about potential issues with each item based on the metrics. Information provided can lead to discussions about how the test items' results might be interpreted or about potential reasons to alter or remove the itms from future administrations of the same test.
Important note: a number from a statistical function calculated for an item does not supplant the importance of practitioner validity. Test authors choose or design items on an assessment with intent, informed by their own expertise about the curriculum being measured. This practice embeds validity into the assessment creation; the conclusions the test authors seek to make about student learning drove their choices. Item statistics included in this tool are meant to supplement practitioner-embedded validity.
Below is information about each piece of information available in this section for each item. Each column is sortable (except for response columns).
Standards and Process Standards
This is the statistics term for Item Difficulty. Item Difficulty on a local assessment is essentially the proportion of students that got it correct. (This definition is slightly modified for constructed response items graded on a rubric scale.) Possible scores for p-value range from 0-1.
Since most (probably all) local assessments are criterion-referenced tests, the intent is for students to demonstrate learning by getting many items correct. For this reason, p-values should not be very low on very many items. It is sometimes desired to have some more difficult items, as well as some less difficult ones. The Local Test Analytics tool will flag items with p-values >= 0.95 as "easy" and <= 0.15 as "hard".
Item Total Correlation
This number represents the relationship between the item's performance and the total performance on the assessment. It is often referred to as "Item Discrimination" and uses a point-biserial function to be calculated. The concept is that an item's performance should discriminate among students based on how well they each score on the whole test. A student with a high score should do well on the item, while a student who scores lower should be more likely to miss the item. Possible scores range from -1 to 1, but negative values are rare and suggest that an item may have a faulty key (high scoring students missed it while low scoring students did not).
Desired point-biserial values for a typical criterion-referenced test consisting of mostly multiple choice items range from 0.2 to 0.39. While the formula allows for values to approach 1, it is highly unlikely that any item on an Aware local assessment will be that high. The Local Test Analytics tool will flag an item if it has a value for Corrected Item Total Correlation lower than 0.15 (with special attention given to negative values). See below for a better definition of "corrected" discrimination.
Corrected Item Total Correlation
The Local Test Analytics tool provides a corrected correlation for item discrimination. This is often a preferred indication of this value because it relates the item's performance to the performance of the test excluding the item itself.
Reliability If Deleted
Will this test have a higher reliability coefficient if this item were not on it? A test author may want to consider removing an item on a future administration of a test if the item does not add to the overall reliability of the assessment.
These results should match what users see in the Aware Analyze tab when viewing data. They can lead to distractor analysis. Just as the correct answer should be always correct, each distractor should be always incorrect. Analyzing distractors can lead to pedagogical discussions about student learning, but they may also reveal weaknesses in the response options.