Usage
  • 52 views
  • 24 downloads

The effect of person misfit on item parameter estimation: A simulation study

  • Author / Creator
    Mousavi, Seyed Amin
  • The validity and reliability of a test may be compromised because of the presence of misfitting response patterns in test data. Consequently, the purpose of the present study was to investigate the effects of the inclusion and exclusion of misfitting response patterns on item parameter estimates under using simulated data. The four factors considered included test length (20, 40, and 60 items), item parameter estimation method (MLE and Baysian Modal), percentage of students who responded misfittingly in the sample (10%, 20%, and 30%), and percentage of items susceptible to misfitting responses (25% and 50%). Two person fit indices (lz and HT) were used to remove misfitting response patterns from the data set with misfitting response patterns. The 2-PL IRT model used to analyze the data. The dependent variables were the bias in the estimated b and a parameters, the standard error of the estimated parameters, and the classification accuracy of placing students into one of two performance categories. The results showed that 1) there was no difference between the two item parameter estimators, 2) the item difficulty parameter (b) was less affected by the presence of misfitting response patterns than the item discrimination parameter (a), 3) item parameters with large true b and a parameter values were affected, 4) an increase in the percentage of misfitting response patterns led to larger bias for both the b and a parameters, 5) the standard errors of estimates of the b and a parameters were small across all conditions, 6) the classification accuracy was higher for the low performing students for the fitting data set and lower and essentially constant across the data sets with all the misfitting response patterns, and the data sets with misfitting response patterns removed by lz and by HT, 7) the classification accuracy was lower for the high performing students than the low performing students for the fitting data set and lower and essentially constant for the remaining three data sets. Implications for practice and recommendations for future research are provided.

  • Subjects / Keywords
  • Graduation date
    2015-11
  • Type of Item
    Thesis
  • Degree
    Doctor of Philosophy
  • DOI
    https://doi.org/10.7939/R3RV0D58C
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.
  • Language
    English
  • Institution
    University of Alberta
  • Degree level
    Doctoral
  • Department
    • Department of Educational Psychology
  • Specialization
    • Measurement,Evaluation and Cognition
  • Supervisor / co-supervisor and their department(s)
    • Ying Cui (Educational Psychology)
  • Examining committee members and their departments
    • Todd Rogers (Educational Psychology)
    • Ruth Childs (Leadership, Higher and Adult Education, University of Toronto)
    • Ken Cor (Pharmacy and Pharmaceutical Sciences)
    • Michael Carbonaro (Educational Psychology)
    • George Buck (Educational Psychology)