DIFFERENTIAL ITEM FUNCTIONING AND ADVERSE IMPACT: A COMPARISON OF MANTEL-HAENZSEL AND LOGISTIC REGRESSION Heather JoAn Whiteman B.A., University of California, Davis 2006 THESIS Submitted in partial satisfaction of the requirements for the degree of MASTER OF ARTS m PSYCHOLOGY (Industrial/Organizational Psychology) at CALIFORNIA STATE UNNERSITY, SACRAMENTO SPRING 2011 DIFFERENTIAL ITEM FUNCTIONING AND ADVERSE IMPA CT: A COMPARISON OF MANTEL-HAENZSEL AND LOGISTIC REGRESSION A Thesis by Heather JoAn Whiteman Approved by: ittee Chair / S. Meyers, P ,SecondR~ Phli Luerrlian, 11 Student: Heather JoAn Whiteman I certify that this student has met the requirements for format contained in the University format manual, and that this thesis is suitable for shelving in the Library and credit is to be awarded for the thesis. (ij'r:>/~1.; · I Date Department of Psychology iii Abstract of DIFFERENTIAL ITEM FUNCTIONING AND ADVERSE IMPA CT: A COMPARISON OF MANTEL-HAENZSEL AND LOGISTIC REGRESSION by Heather JoAn Whiteman This study serves as a comparative analysis of two measures for detecting differential item functioning (DIF) in item responses of 29,171 applicants on a 49 item selection test. The methods compared in this study were two of the more commonly used DIF detection · procedures in the testing arena: the Mantel-Haenszel chi-square and the logistic regression procedure. The study focused on the overall effect each method had on adverse impact when used for the removal of items from a test. The study found that the presence of adverse impact findings were decreased by the removal of items displaying DIF, and that the effect on adverse impact differed by method ofDIF detection. The study does not however, provide e~ough evidence to support the use of one DIF detection method over the -other in applied settings where considerations such as cost and test reliability are of concern. rittee Cbak P . Lawrence S. Meyers, ACKNOWLEDGMENTS I would first like to thank Biddle Consulting Group, Inc. the equal employment opportunity, affirmative action, and employee selection firm in the western United States who allowed me to use their data. I would like to thank all of the professors in the Industrial/Organizational Psychology program who have.influenced me. I would particularly like to thank Dr. Meyers, Dr. Kuthy, and Dr. Berrigan who served on my committee and who assisted in guiding me through the thesis process. I would like to especially appreciate and thank Dr. Meyers for the time he took in giving me advice on the thesis and the enthusiasm and exceptional teaching he offered in his courses. I attribute the bulk of my learning in Industrial/Organizational Psychology to Dr. Meyers and the courses that he taught. I would like to thank Dr. Kuthy for serving as a professional role model and instilling in me a knowledge and respect for the applied Industrial/Organizational field. I would like to thank Dr. Berrigan for his dedication to the students and his willingness to contribute to my thesis. I would also like to thank my parents, Randy and Carrie, who have always been supportive of my goals. I would also like to thank my friends and colleagues who have provided support, input and camaraderie. v TABLE OF CONTENTS Page Acknowledgments ................................................................................... v List of Tables ........................................................................................ ix List of Figures .......................................... :. ........................................... xi Chapter 1. INTRODUCTION ................................................................................ 1 Early History of Selection Testing ........................................................ 1 Setting Legal Precedent for Fairness in Selection Testing ............................ 4 Setting the Standards for Fairness in Selection Procedures ........................... 7 Adverse Impact .............................................................................. 9 Validity in Selection Procedures ......................................................... 13 Evidence Based on Test Content ............................................... 14 Evidence Based on Relation of a Test to Other Variables ................... 16 Evidence Based on Response Processes ....................................... 19 Evidence Based on Internal Structure of a Test ............................... 20 Evidence Based on Consequences of Testing ................................ 22 Validity and Reliability ................................................................... 22 Differential Item Functioning ... : ........................................................ 2.6 Differential Item Functioning and Item Bias ................................. 26 Measuring Differential Item Functioning .................................... 29 Ability Measures in DIF Analyses .............................................. 30 vi Factors Affecting DIF Detection .............................................. 33 Uniform/N"on-Uniform DIF .................................................... 34 DIF Detection Methods .................................................................. 36 Mantel-Haenszel ................................................................. 36 Logistic Regression .............................................................. 40 DIF Detection Method Comparisons ........................................... 44 Purpose of the Study ........................................................................ 46 2. METHOD ......................................................................................... 48 Sample Description ........................................................................ 48 Instrument .................................................................................. 48 Procedure ................................................................................... 50 DIF Analysis for Item Removal ................................................ 50 Adverse Impact Analyses ....................................................... 54 3. RESULTS ........................................................................................ 56 DIF and Item Removal .................................................................... 56 Mantel-Haenszel Analyses ...................................................... 56 Logistic Regression Analyses .................................................. 61 Comparison of the MH and LR Methods for DIF Detection and Item Removal ............................................. , ....................... 64 Adverse Impact Analyses ................................................................ 70 Original Test 80% Rule Adverse Impact Analyses ......................... 72 MH Test 80% Rule Adverse Impact Analyses .............................. 74 LR Test 80% Rule Adverse Impact Analyses ................................ 77 Vil Comparison of 80% Rule Adverse Impact Analyses ........................ 78 4. DISCUSSION .................................................................................... 86 Findings & Conclusions .................................................................. 86 Limitations ................................................................................. 89 Implications for Future Studies ......................................................... 91 Appendices .............................................................................................. 93 Appendix A. Item Means and Standard Deviations ............................................ 94 Appendix B. MH DIF Values and Classification Level by Item ............................. 96 Appendix C. Nagelkerke R2 Values and DIF Classification Category by Item and Comparison Group ................................................................................................ ·l 06 Appendix D. Number of Applicants Passing at Cut-off Score Level by Test and Comparison Group ........................................................................... 120 Appendix E. Fisher's Exact Statistical Significance Results of Adverse Impact by Test & Comparison Group .................................................................... 125 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . 130 Vlll LIST OF TABLES Page 1. Table 1 Demographic Characteristics of Examinees ................................. 49 2. Table 2 Descriptive Statistics ofExaminee Test Scores ............................. 50 3. Table 3 MH Method DIF Classifications by Reference Group ..................... 57 4. Table 4 MH Method DIF Classifications by Item Number .......................... 59 · 5. Table 5 Item Numbers Displaying Small or No DIF with the MH Method ...... 60 6. Table 6 Descriptive Statistics of the MH Test Scores ............................... 60 7. Table 7 LR Method DIF Classifications by Reference Group ...................... 62 8. Table 8 LR Method DIF Classifications by Item Number ........................... 62 9. Table 9 Item Numbers Displaying Small or No DIF with the LR Method ....... 63 10. Table 10 Descriptive Statistics of LR Test Scores .................................... 64 11. Table 11 MH & LR DIF Classifications by Item Number .......................... 69 12. Table 12 Descriptive Statistics of the Original, MH and LR Test Scores ......... 70 13. Table 13 Number of 80% Rule Violations and Practically Significant 80% Rule Violations in the Original Test by Comparison Groups ....................... 73 14. Table 14 Number of80% Rule Violations and Practically Significant 80% Rule Violations in the Original Test by Cut-off Score Levels ....................... 75 15. Table 15 Number of80% Rule Violations and Practically Significant 80% Rule Violations in the MH Test by Comparison Groups ............................ 76 16. Table 16 Number of 80% Rule Violations and Practically Significant 80% Rule Violations in the MH Test by Cut-off Score Levels ........................... 77 17. Table 17 Number of 80% Rule Violations and Practically Significant 80% Rule Violations in the LR Test by Comparison Groups ............................. 79 ix 18. Table 18 Number of 80% Rule Violations and Practically Significant 80% Rule Violations in the LR Test by Cut-off Score Levels ............................ 80 x
Description: