Using Artificial Neural Network (ANN) Analysis With Multilevel Data

• In Event: Fresh Perspectives on Estimating Complex Multilevel Models: Counts, Mixtures, Interactions, and Machine Learning

Abstract

Our study investigates the performance of Artificial Neural Network (ANN) analysis with multilevel data. ANN is a machine-learning algorithm used for exploratory research with large-scale data. The goal of ANN is to minimize prediction error and identify “important” predictors. Advantages of ANN over regression include that its robustness to non-linear X-Y relations and multicollinearity problems, and its built-in cross-validation procedures. However, prior methodological research has reported conflicting results about the usefulness of ANN for multilevel data. Using simulations, we found that the accuracy of ANN in reproducing the true R2 depended on the size of R2 and ICCs among predictors and outcomes. Final analyses will also report on other predictor patterns and varied numbers of hidden layer (latent) nodes.

Authors

• Emine Yavuz, Erciyes University

• Elizabeth A. Sanders, University of Washington

• Timothy R. Konold, University of Virginia