Handling Missing Data in Variational Autoencoder Based Item Response Theory
Author Information
Author(s): Karel Veldkamp, Raoul Grasman, Dylan Molenaar
Primary Institution: Department of Psychology, University of Amsterdam
Hypothesis
Can variational autoencoders effectively handle missing data in item response theory models?
Conclusion
Variational autoencoder methods are a time-efficient alternative to traditional methods for estimating item response theory models, especially when dealing with missing data.
Supporting Evidence
- Variational autoencoders can estimate high-dimensional item response theory models efficiently.
- Different methods for handling missing data were systematically compared.
- Results showed that variational methods can outperform traditional maximum likelihood methods in certain scenarios.
Takeaway
This study looks at how to fill in missing answers when students take tests, using smart computer methods to make better guesses about what they would have answered.
Methodology
The study adapted and compared four different variational autoencoder methods for handling missing data in item response theory models through simulation studies.
Potential Biases
Potential biases in item parameter estimates due to the handling of missing data.
Limitations
The study assumed that all data were missing completely at random, which may not reflect real-world scenarios.
Participant Demographics
The study involved simulated datasets of 10,000 respondents.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website