🔎 What This Paper Addresses

Right-censored cases in political science survival data are often misclassified as failure events because of measurement error. Treating those cases as real failures underestimates event durations and can bias coefficient estimates—especially when misclassification is related to covariates of interest. A new estimator is introduced to correct this source of bias.

🧭 Model and Estimation Approach

A Bayesian split-population survival model is developed that explicitly models misclassified failure events alongside the parametric survival process of interest. Key features include:

• A two-equation system: one equation for the probability a recorded failure is truly a failure, and one parametric survival equation for the duration process.
• Bayesian estimation implemented via slice sampling to obtain posterior inference for both misclassification parameters and survival parameters.

🧪 How the Model Was Evaluated

• Performance assessed with simulated datasets to examine bias and recovery of true parameters under varying misclassification scenarios.
• Applied to several political science cases, including analyses of civil war duration and democratic survival.

📌 Key Findings

• The split-population Bayesian estimator successfully accounts for misclassified failure events and reduces bias in estimated durations.
• Coefficient estimates that would be distorted when misclassification correlates with covariates are recovered more accurately under the proposed model.
• The method performs well in both simulation and substantive applications (civil war duration; democratic survival).

🌐 Why This Matters

Misclassified failures are a common but under-addressed source of bias in survival analyses within political science. Using an explicit misclassification model with Bayesian estimation improves inference about event duration and the effects of covariates, strengthening conclusions in studies of conflict and regime durability.