Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier


Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program


Year Degree Awarded


Month Degree Awarded


First Advisor

Scott Monroe

Second Advisor

Stephen G. Sireci

Third Advisor

Frederic Robin

Subject Categories

Educational Assessment, Evaluation, and Research


On low-stake tests, unmotivated test takers may show disengaged behaviors, such as rapid responses with short response times and such non-effortful responses can be a threat to the test score validity. Several methods have been proposed to determine rapid-guessing behavior, which can be classified as response time threshold methods and item response theory based methods. In addition, I suggest a machine learning method based on neural networks known as the autoencoder for anomaly detection. This research aims to compare the detection of rapid-guessing using various methods and their effects on inferences drawn from test data. In the simulation study, (a) the classification accuracy of the methods and (b) the effects of motivation filtering, i.e., removing rapid-guesses, on person and item parameter estimates are investigated and compared. (c) The methods are also compared using public data from the Programme for International Student Assessment (PISA). (d) Finally, to better understand rapid-guessing detection methods, multiple variables in PISA process data are investigated using another machine learning method, gradient boosting machine. The results showed variations in the detection of rapid-guessing across different methods. Additionally, the impact of motivation filtering differed among methods, exhibiting interactions with the level of filtering and rapid-guessing patterns. The results of gradient boosting machine suggested response time, item-centered response time, and person-centered response time were the key information to predict rapid-guessing behaviors identified by included RG detection methods.


Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Available for download on Thursday, August 01, 2024