Recent advances in object detection have significantly improved performance for autonomous driving perception; however, understanding how detection errors evolve over time remains critically underexplored. In this work, we introduce a Markovian error propagation framework to quantify temporal consistency across four distinct detection states—correct, misclassified, missed, and false positive. Using sequences from the KITTI Tracking dataset, we model how object detectors transition between these states in consecutive frames, revealing characteristic error patterns and recovery behaviors. Our analysis generates empirical transition matrices, error duration statistics, and state recovery probabilities that provide deeper insights than traditional static metrics. The results demonstrate that errors exhibit strong temporal dependencies, with state transition probabilities varying significantly across object classes and environmental conditions. Notably, we find that missed detections persist for approximately 26 frames on average, while misclassifications self-correct more rapidly (typically within 4 frames). This Markovian perspective provides a principled approach to quantifying detection reliability over time, offering both a diagnostic framework for existing models and design guidance for more temporally coherent perception systems in autonomous driving.