🧭 The Problem: Interdependence in Binary Outcome Models
Binary outcome models are common in social science and economics but become difficult to estimate when data exhibit spatial, temporal, or spatio-temporal autocorrelation. Jointly determined error terms in reduced-form specifications are generally analytically intractable. Simulation-based estimators have been proposed to deal with this, but they (i) are computationally intensive and impractical for the large datasets common in contemporary research, and (ii) rarely address temporal interdependence.
🛠️ A Faster Estimator for Interdependent Binary Data
Introduces analytically tractable pseudo-maximum-likelihood (pseudo-MLE) estimators for latent binary choice models that allow interdependence across space and time. Also proposes an implementation strategy designed to increase computational efficiency considerably.
📈 How This Was Evaluated
- Monte Carlo experiments used to compare the new pseudo-MLE estimators with commonly used alternatives
- Evaluation focused on parameter recovery accuracy and computational cost
🔍 Key Findings
- The pseudo-MLE estimators recover parameter values as well as commonly used simulation-based alternatives
- The proposed implementation requires only a fraction of the computational cost of those alternatives
⚖️ Why It Matters
Provides a practical solution for fast, reliable estimation of binary choice models with spatial, temporal, and spatio-temporal interdependence, removing a major computational barrier for researchers working with sizable interdependent datasets.
