Download presentation
Presentation is loading. Please wait.
Published byKathleen Blankenship Modified over 9 years ago
1
Field Test Analysis Report: SAS Macro and Item/Distractor/DIF Analyses
Prepared by Yi-Hsin Chen, Chunhua Cao, and Stephanie Green College of Education at USF Presented at the meeting of the Central Florida Assessment Collaborative (CFAC) May 20th, 2014, Orlando Florida
2
Agenda of This Presentation
SAS macro for CTT test/item analysis, IRT 2PL model, and Mantel-Haenszel differential item functioning (DIF) analysis Introduction of statistical concepts for test/item development Item Analyses: CTT and IRT Distractor Analysis DIF Analysis
3
SAS macro for test/item, 2PL, DIF analyses
4
SAS Macro Outputs A SAS macro developed for this project
There are six excel outputs Test score statistics Frequencies of options for each item Item analysis statistics Distractor analysis DIF 2PL item parameter Available upon request at
5
Test Score Statistics
6
Frequencies of Options
7
Item Analysis Statistics
8
Item Analysis Statistics
9
Distractor Analysis
10
DIF Analysis
11
Statistical Concepts of Test Scores
12
Sample size N: Sample size 85, 60, 70, 44, 59, 89, 99, 79, . , 100
USED_N: Sample size used for analysis without missing data one missing data USED_N = 9
13
Central Tendency MEAN: Arithmetic average
Most frequently reported measure of central tendency Sum of scores divided by number of scores
14
Test Statistics: Central Tendency
MEDIAN (Q2): the score at the 50th percentile half of the examinees score above median, and half score below median 110 105 100 95 90 Median = / 2 = 97.5 110 105 100 95 90 Median = 100
15
Percentiles Percentile is considered when we consider the percentage of scores that fall below a given point They are very useful for interpreting an individual student’s performance Q1: The score is at the 25th percentile Q1 = 10, indicating 25 percent of the students’ scores below 10 points Q3: The score is at the 75th percentile
16
Variability Range Subtract lowest score (Minimum) from highest score (Maximum) This is a rough measure of variability High score = 90 Low score = 50 Range = ? (40) High score = 100 Low score = 50 Range = ? (50) High score = 90 Low score = 30 Range = ? (60)
17
Variability Standard Deviation (SD):
an average points that deviates from the mean score A measure of the amount of variability in examinees’ total scores Large SD = large variability (heterogeneity) Small SD = small variability (homogeneity) (scores cluster closer to the mean)
18
Variability Deviation Scores Squared 100-92= 8 82 = 64
100-92= = 64 = = 16 = = 4 = = 0 = -2 (-2)2 = 4 = -12 (-12)2 = 144 232 = (X-Mean)2 SD = (X-Mean) = = N Scores 100 96 94 92 90 80 Mean = 92 6.22
19
Skewness and Kurtosis SKEWNESS:
a measure to tell the shape of the score distribution, such as positive or negative skewness or symmetry KURTOSIS: a measure of the "peakedness" of the score distribution
20
Skewness
21
Skewness a roughly negatively skewed distribution (bar chart)
22
Skewness
23
Skewness a roughly positively skewed distribution (bar chart)
24
Kurtosis Different kurtosis values K > 0 K = 0 K < 0
25
Reliability: Cronbach’s Alpha
A measure of the test reliability, indicating the internal consistency of the test Sample dependent Different samples may obtain different reliability with the same test Ranges from 0 to 1 0.7 and above: good internal consistency
26
Standard Error of Measurement
SEM (Standard Error of Measurement) SEM = STD * 1−𝑟𝑒𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 A higher reliable test can cause smaller SEM
27
Statistical Concepts of Item Analysis
28
Item Analysis Why care? –
Item analysis helps you identify problems with your items (or scoring) These problems can be corrected, resulting in a better test, and better measurement
29
Item Analysis When is it useful? –
Item analysis is most useful when you are developing a bank, or pool, of items that you will continue to use It can be used when evaluating standardized tests It is also a useful tool, anytime students have complained about an item It can be used to identify mis-keyed items
30
Item Difficulty (p-value)
Item difficulty (proportion correct): the proportion of examinees tested that answered the item correctly # of students who responded correctly total # of students who responded p = Ncorrect Ntotal p =
31
Item Difficulty (p-value)
p can range from 0 to 1.0 A rough level of item difficulty (p) .80 and above moderately easy to very easy (mastery) moderate .30 and below moderately difficult to very difficult
32
Item Discrimination Discrimination can be computed using correlation
This shows the relationship between a single item and the total test It is expected that students with high scores answer the item correctly rpb = (point-biserial) correlation between item score and total score
33
Item Discrimination Corrected point-biserial correlation:
A statistic similar to point-biserial correlations The score of the individual item is taken out of the total score so that the contribution of the item itself is removed from the correlation This statistic is more accurate to represent item discrimination
34
Item Discrimination Two ability groups (upper and lower) approach
Median score is used to divide the students into two groups Discrimination coefficient (D-value) = percentage correct in the upper group – percentage correct in the lower group Ranges from -1 to 1 An item with higher and positive D-value indicates a good discriminating item An item with a negative D-value suggests that the lower achieving group did better on an item than the higher achieving group, indicating a poor item
35
Item Discrimination A rough scale of item discrimination (D)
D can range from -1 to 1 .30 and above moderate to high discrimination little to no discrimination 0 and below negative discrimination (unwanted)
36
Item Difficulty and Discrimination
Relationship between item difficulty and discrimination there can be little discrimination: if nearly everyone gets the item right, or if nearly everyone gets the item wrong there can be maximum discrimination: if about half the people got the item right, and about half got the item wrong
37
Item Difficulty and Discrimination
Item Difficulty Max Discrimination Relationship between item difficulty and potential discrimination
38
Alpha If an Item Deleted
“The Alpha If Deleted” shows what would happen to the internal consistency when the item is deleted When the test_alpha_deleted coefficient goes up, compared with the original test-alpha, it indicates that without the deleted item, the test can be more reliable (that item can be removed from the test) When the test_alpha_deleted coefficient goes down, it means that deleting that item is not a good thing and also indicates that item is a good item
39
Statistical Concepts of Distractor Analysis
40
Distractor Analysis used to determine which distractors students find attractive consider the proportion of (total) students choosing each option compare the number of examinees selecting each option in the High and Low groups, or Example: Proportion of total examinees selecting each option a* b c d Total .78 .11 .03 .08
41
Selecting upper and lower groups
Upper and Lower groups are needed: to hand-compute D-values, and for distractor analysis when comparing numbers of examinees To select Upper and Lower groups: arrange the tests by total score separate out the tests for each group top half becomes Upper group, and bottom half becomes Lower group
42
Selecting upper and lower groups
Upper and Lower groups are needed: to hand-compute D-values, and for distractor analysis when comparing number of examinees To select Upper and Lower groups: Upper group: top half (50%) or top 33% Lower group: bottom half (50%) or bottom 33%
43
Example 1: distractor analysis
1. The capital of Switzerland is Bern. Zurich. Lucerne. Geneva. Numbers in the High and Low groups who selected each option a* b c d Upper 13 1 Lower 3 2 9
44
Example 2: distractor analysis
2. The most important part of test planning is creating: sound instruction. a test blueprint. an item analysis plan. the grading curve. Numbers in the High and Low groups who selected each option a b* c d Upper 1 8 Lower 2
45
Example 3: distractor analysis
3. Which type of essay item contains the most explicit instructions to students? extended response fixed response explicit response restricted response Numbers in the High and Low groups who selected each option a b c* d Upper 3 1 2 14 Lower 4 7 8
46
Statistical Concepts of 2PL IRT model Analysis
47
Two-Parameter Logistic Model
Alpha represents item discrimination The value is positive Beta represents item difficulty with the mean of 0 and the SD of 1 Items with the negative values = easy items Items with the positive values = hard items
48
Statistical Concepts of DIF Analysis
49
Differential Item Functioning
A major concern regarding using the psychological measures is that these measures may “work differently” or be either “for or against” a particular group of examinees (e.g., gender or ethnicity) When a test item unfairly favors one group over another, it can be said to show differential item functioning or DIF
50
Uniform or consistent DIF
51
Non-uniform or crossing DIF
52
Mantel Haenszel chi-square
1 Total Reference Bt At NRt Focal Dt Ct NFt M0t M1t Tt subscript t = individual raw score
53
Mantel Haenszel chi-square
Controlling for the observed score, we want to see if the proportion correct for the focal group is equal to that for the reference group on an item The MH statistic consists of a series of 2x2 contingency tables MH = 1 : No DIF MH < 1: DIF and favor the focal group (dummy=0) if p < .05 MH > 1: DIF and favor the reference group (dummy=1) if p < .05
54
Field Test Analyses
55
Test Statistics for Three Subjects
Precalculus N 210 USED_N MEAN 9.748 STD 2.978 MIN 2 Q1 8 MEDIAN 10 Q3 11 MAX 20 SKEWNESS 0.378 KURTOSIS 0.679 ALPHA 0.506 SEM 2.093 STATISTIC Anatomy N 269 USED_N MEAN 12.364 STD 3.337 MIN 4 Q1 10 MEDIAN 12 Q3 15 MAX 21 SKEWNESS -0.102 KURTOSIS -0.447 ALPHA 0.533 SEM 2.281 STATISTIC Phy-Sci N 183 USED_N MEAN 12.852 STD 4.141 MIN 4 Q1 10 MEDIAN 13 Q3 16 MAX 25 SKEWNESS 0.088 KURTOSIS -0.658 ALPHA 0.626 SEM 2.531
56
Item Difficulty Item difficulty Pre-calculus (21 items) 0-0.1
(2 Items) 19, 3 (1 Item) 14, 1 (14 Items) 21, 10, 11, 18, 12, 20, 8, 16, 17, 15, 14, 6, 13, 2 (3 Items) 7, 5, 9 Item difficulty Anatomy (27 items) 0-0.1 (0 items) (2 items) 13, 2 (6 items) 16, 8, 27, 3, 10, 20 (14 items) 17, 4, 9, 14, 11, 5, 18, 26, 25, 7, 15, 22, 21, 24 (3 items) 12, 19, 1 (2 items) 23, 6 Item difficulty Physical Science (31 items) 0-0.10 (1 item) 22 (2 items) 11, 28 (5 items) 16, 27, 9, 6, 20 (12 items) 30, 18, 12, 25, 31, 15, 2, 19, 24, 13, 26, 29, 21, 10, 23, 8, 7, 4, 3, 17, 5, 14 (1 item) 1 0 items
57
Item Discrimination (Corrected point-biserial correlation)
Value Pre-calculus 21 items Negative Value (1 Item) 19 0-0.10 (2 Items) 11, 3 (13 Items) 9, 10, 2, 18, 8, 5, 17, 21, 1, 13, 14, 4, 12 (5 Items) 16, 15, 7, 6, 20 Above 0.30 (0 Items) Value Pre-calculus 27 items Negative Value (3 items) 3, 17, 13 0-0.10 (6 items) 16, 9, 20, 10, 27, 2 (9 items) 15, 4, 11, 26, 18, 12, 5, 14, 8 (9 items) 7, 1, 25, 22, 23, 21, 24, 6, 19 Above 0.30 0 items Value Physical Science 31 items Negative Value (6 items) 11, 22, 20, 12, 31, 1 0-0.10 (2 items) 21, 5 (8 items) 23, 19, 6, 28, 25, 18, 10, 16 (6 items) 8, 3, 15, 2, 30, 27 Above 0.30 (9 items) 13, 7, 17, 24, 29, 9, 14, 26, 4
58
Item Discrimination (Two-Group Approach)
Value Pre-calculus 21 items Negative Value (0 Items) 0-0.10 (3 Items) 19, 3, 18 (7 Items) 21, 14, 9, 17, 10, 2, 11 (7 Items) 1, 5, 12, 13, 8, 7, 16 Above 0.30 (4 Items) 15, 20, 4, 6 Value Anatomy 27 items Negative Value (0 items) 0-0.10 (5 items) 13, 3, 16, 17,2 (7 items) 27, 9, 10, 20, 12, 6, 26 (11 items) 8, 4, 11, 1, 15, 18, 23, 15, 5, 7 Above 0.30 (5 items) 25, 24, 19, 22, 21 Value Number of items Negative Value (2 items) 11, 22 0-0.10 (4 items) 20, 1, 12, 31 (8 items) 28, 21, 23, 5, 19, 25, 6, 16 (6 items) 10, 3, 8, 27, 18, 9 Above 0.30 (10 items) 15, 30, 2, 17, 13, 24, 14, 7, 29, 26, 4
59
Alpha Difference (Alpha and Alpha When deleted)
Pre-Calculus 21 items Negative Value (2 Items) 19, 11 (3 Items) 3, 9, 10 (5 Items) 2, 8, 18, 5, 17 Above 0.01 (14 Items) 21, 1, 13, 14, 4, 12, 16, 7, 15, 6, 20 Alpha Difference Anatomy 27 items Negative Value (7 items) 3, 17, 9, 16, 13, 20, 10 (3 items) 27, 2, 15 (4 items) 4, 11, 26, 18 Above 0.01 (13 items) 12, 5, 14, 8, 1, 7, 25, 23, 6, 22, 21, 19, 24 Alpha Difference Physical Science 31 items Negative Value (8 items) 11, 20, 12, 31, 22, 1, 21, 5 (6 items) 23, 19, 6, 28, 25, 18 (3 items) 10, 16, 8 Above 0.01 (14 items) 3, 15, 2, 27, 30, 13, 7, 17, 24, 9, 29, 14, 26, 4
60
Item Analysis Summary The test with reliability (alpha) less than .5 needs to be worried Too hard item (e.g., p-value < 0.1 or 0.2) or/and too easy (e.g., p-value close to 1) items may be revisited Revisiting Items with a negative value of discrimination is warranted, especially for the two-group item discrimination Items with negative alpha difference between the original test alpha and the test alpha when deleted are not good, either
61
DIF Results: Precalculus
Girls = 0 Boys = 1 Favor boys
62
DIF Results: Precalculus
Girls = 0 Boys = 1 Favor girls
63
DIF Results: Anatomy Girls = 0 Boys = 1 Favor boys
64
DIF Results: Anatomy Girls = 0 Boys = 1 Favor girls
65
DIF Results: Anatomy Girls = 0 Boys = 1 Favor boys
66
DIF Results: Anatomy Girls = 0 Boys = 1 Favor girls
67
DIF Results: Anatomy Girls = 0 Boys = 1 Favor girls
68
DIF Results: Physical Science
Girls = 0 Boys = 1 Favor boys
69
DIF Results: Physical Science
Girls = 0 Boys = 1 Favor boys
70
DIF Results: Physical Science
Girls = 0 Boys = 1 Favor girls
71
Distractor Analysis: Typical Problems and Solutions
72
Precalculus: Item 29 Frequency Row Pct Table of groupB by r19 groupB
D Total LOWER GROUP 5 9.80 51 UPPER GROUP 8 7.02 114 61 53 38 13 165 Frequency Missing = 76
73
Precalculus: Item 29 The item is a hard item (p = 0.18)
74
Precalculus: Item 3 Frequency Row Pct Table of groupB by r3 groupB r3
Table of groupB by r3 groupB r3 A B C* D Total LOWER GROUP 4 6.45 62 UPPER GROUP 6 4.55 132 60 85 39 10 194 Frequency Missing = 47 Add items 3 and 1 for worse items Add item 21 as a good item with high discrimination and high difficulty Add two items with moderate difficulty and high discrimination The item is a hard item (p = 0.162)
75
Precalculus: Item 3 The item is a hard item (p = 0.19)
Add items 3 and 1 for worse items Add item 21 as a good item with high discrimination and high difficulty Add two items with moderate difficulty and high discrimination The item is a hard item (p = 0.19)
76
Precalculus: Item 1 Frequency Row Pct Table of groupB by r1 groupB r1
Table of groupB by r1 groupB r1 A B C D* Total LOWER GROUP 5 8.62 58 UPPER GROUP 8 6.20 129 37 76 13 61 187 Frequency Missing = 54 The item is a hard item (p = 0.253)
77
Precalculus: Item 1 The item is a hard item (p = 0.30)
78
Precalculus: Item 14 Table of groupB by r14 groupB r14 - A B C D*
Total LOWER GROUP 5 5.56 8 8.89 3 3.33 90 UPPER GROUP 5 4.55 4 3.64 110 10 12 7 116 55 200 Frequency Missing = 10
79
Precalculus: Item 14 The item is challenging (p = 0.26)
Option C may be the potential key Or students have a misconception on this item
80
Precalculus: Good Item
Table of groupB by r21 groupB r21 - A B* C D Total LOWER GROUP 99 UPPER GROUP 111 56 34 66 28 26 210 The item is challenging (p = 0.266) Discriminating well
81
Precalculus: Good Item
The item is challenging (p = 0.31) Discriminating well However, this item shows DIF and favors girls
82
Summary for Precalculus
Some items need to revisit: Items: 19, 3, 1, and 14 Develop some easy items (p= ) Two DIF items Items 4 and 21
83
Anatomy: Hard Item The item is a hard item (p = 0.271)
Table of groupB by r3 groupB r3 A B C* D Total LOWER GROUP 108 UPPER GROUP 7 4.38 160 27 52 73 116 268 Frequency Missing = 1 The item is a hard item (p = 0.271) Not discriminating well
84
Anatomy: Hard Item The item is a hard item (p = 0.271)
Not discriminating well
85
Anatomy: Potential Miskey
Table of groupB by r16 groupB r16 A B C D* Total LOWER GROUP 108 UPPER GROUP 160 144 43 22 59 268 Frequency Missing = 1 The item may have a miskey of Option D The possible correct key is Option A (Majority of the upper group chose this option)
86
Anatomy: Potential Miskey
The item may have a miskey of Option D The possible correct key is Option A (Majority of the upper group chose this option) Or there is a misconception on this item
87
Anatomy: Good Item The item has moderate difficulty level(p = 0.491)
Table of groupB by r25 groupB r25 A B C D* Total LOWER GROUP 109 UPPER GROUP 8 5.03 159 21 52 63 132 268 Frequency Missing = 1 The item has moderate difficulty level(p = 0.491) Discriminating well
88
Anatomy: Good Item The item has moderate difficulty level(p = 0.491)
Discriminating well
89
Summary for Anatomy The p-value of the items look good, with half of the items being moderate difficult, almost one quarter of them being easy, and almost one quarter being difficulty No negative discrimination items using the two-group approach (a good sign) The test alpha is low (0.533) DIF: Items 14, 19 (favoring boys) and items15, 22, 26 (favoring girls)
90
Physical Science: Item too hard
Table of groupB by r28 groupB r28 - A B C D* Total LOWER GROUP 2 2.33 86 UPPER GROUP 4 4.12 8 8.25 97 22 23 81 27 30 183 The item is a hard item (p = 0.164)
91
Physical Science: Item too hard
The item is a hard item (p = 0.164)
92
Physical Science: Potential Miskey
Table of groupB by r11 groupB r11 A B C* D Total LOWER GROUP 8 9.30 86 UPPER GROUP 9 9.28 97 101 37 28 17 183 The item may have a miskey of Option C The possible correct key is Option A (Majority of the upper group chose this option)
93
Physical Science: Potential Miskey
94
Physical Science: Good Item
Table of groupB by r27 groupB r27 A B C D* Total LOWER GROUP 86 UPPER GROUP 9 9.28 97 22 49 67 45 183 The item has moderate difficulty level(p = 0.491) Discriminating well
95
Physical Science: Good Item
96
Summary for Physical Science
Some items need to revisit: Items: 6, 11, 12, 22 Potential miskey item: 11 Develop some easy items (p= ) DIF: Items 3 and 4 (favoring boys) and Item 7 (favoring girls)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.