Partial Credit Scoring for Technology Enhanced Items - PowerPoint Presentation

Partial Credit Scoring for Technology Enhanced Items CCSSO National Conference on Student Assessment 22 June 2016 1

Overview Possible scoring rules for technology enhanced items (TEIs) DichotomousPartial credit scoring (polytomous)Evaluating whether partial credit scoring is “working”Study methods & resultsAdditional considerationsDo TEIs with more correct answers take longer to answer?Are TEIs with more correct answers more difficult?Do TEIs take more time to answer compared to multiple-choice items?Should TEIs be worth more points than multiple-choice items?How might partial credit scoring impact student subgroups? 2

Possible Scoring Rules for TEIs Dichotomous Currently used in VirginiaStudents must select all correct answers to receive 1 point; all other responses receive 0 points PolytomousVirginia is looking to implement partial credit scoring in the future for TEIsStudents get “full credit” if they select all correct answers Different methods allow for different ways to determine how partial credit versus no credit is determinedThis research looked at three possible methods:“N Method”“N-1 Method”“N/2 Method”3

Partial Credit Scoring: N Method N = the total number of correct answers for a particular TEI Each correct response receives partial credit For example, if there are 5 correct answers, each correct answer could be worth 1 point for a total of 5 possible points Items with different numbers of correct answers will have different numbers of score pointsCould scale each item to have the same score range (e.g., 0-1), but there will be N+1 score categories (e.g., 0, 1/5, 2/5, 3/5, 4/5, and 1)4

Partial Credit Scoring: N Method Benefits: perceived face validity might be similar to how credit would be awarded in the classroom for multi-part items Limitations: each score point may not really be discriminating between distinct levels of content knowledgemay be difficult to obtain sufficient numbers of students within all score categorieshaving items with different score ranges may make meeting a test blueprint more challenging5

Partial Credit Scoring: N-1 Method N = the total number of correct answers for a particular TEI “N-1 Method” – Students get partial credit if they select all but one correct answer; if they select two or more incorrect answers, they receive no credit N-1 Method will have 3 score categories for all items (e.g., 0, 1, 2) Example: If an item has 5 correct answers and the student selects all 5 correct, they receive full credit. If the student selects 4 of the 5 correct answers, they receive partial credit, and if they get 0-3 correct answers, they receive no credit.6

Partial Credit Scoring: N-1 Method Benefits: all items, regardless of the number of correct answers, have three possible score categories having the same score range for all TEIs will make meeting a test blueprint more straight-forward with only three score categories, sparse data is less likely to be a problem, which will make parameter estimation more stableLimitations: one-size-fits-all scoring approach may not work well for very different TEI typestends to be a very conservative partial credit scoring approach such that very few students may obtain partial credit (dichotomous outcome)7

Partial Credit Scoring: N/2 Method N = the total number of correct answers for a particular TEI “ N/2 Method” – Students get partial credit if they select at least half of the correct answers; if they select less than half of the correct answers, they receive no credit N/2 Method will have 3 score categories for all items (e.g., 0, 1, 2)N/2 Method and N Method are identical for items with 2 correct answersExample: If an item has 5 correct answers and the student selects all 5 correct, they receive full credit. If the student selects 3 or 4 of the 5 correct answers, they receive partial credit, and if they get 0-2 correct answers, they receive no credit.8

Partial Credit Scoring: N/2 Method Benefits: all items, regardless of the number of correct answers, have three possible score categories having the same score range for all TEIs will make meeting a test blueprint more straight-forward with only three score categories, sparse data is less likely to be a problem, which will make parameter estimation more stableLimitations: one-size-fits-all scoring approach may not work well for very different TEI types9

Is PCS “Working”? Evaluation Criteria Number of Students in Each Score Category Do we have enough data to estimate differences in difficulty from one category to the next?Are we making a meaningful differentiation between groups of students? As item scores increase, is the total test score increasing? Are item-total correlations increasing ?Are we improving measurement precision, or just adding “noise” into student scores?10

Methods Focused on newly field-tested TEIs in one grade of science and one grade of mathematicsAnalysis focused on Hot Spot and Drag and Drop TEIs that required two or more correct answers for full credit. Note: This study used items that were not specifically developed with partial credit scoring in mind. Virginia intends to begin development of TEIs that are created with partial credit scoring considered from the outset moving forward. 11

Results 12 Item Type Number of Students Scoring Approach Dichotomous N/2 0 1 0 1 2 Hot Spot 5,231 31.27 39.07 26.91 34.00 39.07 3,938 26.42 35.99 23.68 26.64 35.99 3,904 32.51 35.61 21.83 32.62 35.61 Drag-and-Drop 3,939 29.34 35.38 21.13 29.53 35.38 3,95530.6237.5930.2231.0837.595,08925.4335.7217.6226.7735.72 Item Type Number of Students Scoring Approach N 0 1 2 3 4 5 Hot Spot 5,231 24.50 27.40 34.00 39.07     3,938 23.68 26.64 35.99       3,904 13.00 – 22.64 27.76 33.00 35.61 Drag-and-Drop 3,939 – 25.25 17.00 29.53 – 35.38 3,955 31.26 23.11 30.71 31.54 37.59   5,089 17.50 17.71 26.42 27.25 35.72  

Results 13 Item Type Scoring Approach Dichotomous N/2 N Hot Spot 0.42 0.51 0.51 0.35 0.35 0.35 0.17 0.17 0.19 Drag-and-Drop 0.20 0.20 0.20 0.39 0.38 0.35 0.34 0.35 0.35

Results N Method Many score categories with very few students, especially for TEIs with a larger numbers of correct answers Many cases where the total score did not increase as the item score increased N-1 Method Very few students would receive partial credit using this method N/2 MethodMost consistent with how content experts assigned partial credit scoring to TEIsAppears to work well with Virginia items in terms of having sufficient numbers of students within each of the three score categoriesDid not appear to result in huge improvements in measurement properties (item total correlations), but also did not systematically decrease technical quality of scores14

Do TEIs with More Correct Answers Take Longer to Answer? 15

Are TEIs with More Correct Answers More Difficult? 16

Do TEIs take More Time than MC Items to Answer? 17 Subject TEI MC Min Max Average Average Science 28.32 214.73 81.45 58.50 Math 76.49 466.98 185.55 132.29

Should TEIs be Worth More Points than MC Items? Yes ? On average TEIs take more time to answer TEIs often require more than one student interaction, while MC items require students to select only one answerTEIs often used to try to measure higher level skillsTEIs look similar to items that might receive more than a 0/1 score in the classroom18

Should TEIs be Worth More Points than MC Items? No? Number of interactions not correlated with the amount of time students spend on an itemTEIs not necessarily more difficult than MC items when scored dichotomouslyOften each interaction is not measuring a distinct skill, but is just more thoroughly evaluating a single skill19

How Might Partial Credit Scoring Impact Student Subgroups? TEIs make up 15-20% of the items on Virginia assessments Partial credit scoring essentially splits the category of students who received 0 points under dichotomous scoring into 2 or more score categories—no credit and various degrees of partial creditThis may help lower performing students show some content mastery However, each type and subtype of TEI often requires understanding of a different mechanism for responding (e.g., various forms of clicking and/or dragging)This may result in an additional obstacle for English Language Learners and Poor Readers who may struggle to understand item-specific response directionsThe Virginia Department of Education provides practice items and released items to give students and schools an opportunity to become familiar with different types of TEIs20

Take-Aways Not all technology enhanced items support partial credit scoring (e.g., one blank FIB items) For items that do support partial credit scoring, determining which responses merit partial credit is not always straight-forward There are many methods for applying partial credit, some which may work better than others for a given context Partial credit scoring may not result in vast improvements in measurement precision but may still enhance face validity and public perception21