Related Experiment Videos
The effect of dropping low scores on ability estimates
1Department of Psychology, University of Virginia, P. O. Box 400400, Charlottesville, VA 22904-4400, USA. rpbowles@virginia.edu
Journal of Applied Measurement
|April 6, 2004
Summary
Dropping low scores in assessments is common but not supported by research. Simulations show that using all scores, rather than dropping low ones, provides more valid ability estimates.
Area of Science:
- Psychometrics
- Educational Measurement
- Statistical Modeling
Background:
- Dropping low scores is a prevalent method for combining results from multiple assessments.
- The psychometric validity of ability estimates derived from dropping low scores remains unexamined.
Purpose of the Study:
- To investigate the validity of ability estimates when low scores are dropped.
- To compare the impact of dropping low scores on Rasch estimation and proportion correct scoring methods.
Main Methods:
- A simulation approach was employed to estimate bias, root mean squared error (RMSE), and examinee benefit.
- Simulations were conducted under three conditions: normal, bad day, and bad test.
Main Results:
- Ability estimates using complete data were generally superior to those obtained after dropping low scores.
- The practice of dropping low scores was found to be unwarranted in most assessment scenarios.
Conclusions:
- The findings challenge the conventional practice of dropping low scores in score aggregation.
- Assessment practitioners should exercise caution when employing the 'dropping low scores' technique due to potential validity issues.