Transformers deep learning models for missing data imputation: an application of the ReMasker model on a psychometric scale
2024

Using Transformers for Missing Data Imputation in Psychology

Sample size: 15740 publication Evidence: high

Author Information

Author(s): Casella Monica, Milano Nicola, Dolce Pasquale, Marocco Davide

Primary Institution: University of Naples “Federico II”

Hypothesis

Can transformer-based deep learning models effectively impute missing data in psychometric research?

Conclusion

The ReMasker model outperforms traditional imputation techniques and shows comparable performance to other machine learning methods.

Supporting Evidence

  • The ReMasker model consistently provided the most accurate imputations across all tested scenarios.
  • Machine learning techniques, particularly ReMasker, achieve superior performance in terms of reconstruction error.
  • Conventional methods showed the least favorable performance compared to machine learning techniques.

Takeaway

This study shows that a new computer model can fill in missing information in psychological surveys better than older methods.

Methodology

The study compared the ReMasker model with traditional and machine learning imputation methods using a psychometric dataset.

Potential Biases

The reliance on specific assumptions about missing data mechanisms may introduce bias.

Limitations

The study focused only on continuous data imputation and did not explore all possible imputation techniques.

Participant Demographics

Responses from 15,740 individuals aged 18 and older from 137 countries.

Digital Object Identifier (DOI)

10.3389/fpsyg.2024.1449272

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication