Does NYT really have the vote counts time series available publicly? It would be an interesting dataset for future study. You either prove there is anomaly (for various possible reasons) or you find ways to model it and explain it rationally.