Applying statistical and causal modeling to interpret thermal alteration in observational lithic data

Abstract

Wildfire events are increasingly recognized as agents of taphonomic alteration in archaeological contexts. This study applies causal and statistical modeling to assess the impact of thermal alteration on lithic assemblages in the San Matías Gulf region of Northern Patagonia, Argentina. Utilizing data derived from naturalistic sampling after recent wildfires, we developed a Directed Acyclic Graph (DAG) to identify key mediators and potential confounders affecting lithic fragmentation. A minimal adjustment set was determined to isolate the indirect effect of fire on mechanical alterations, and Random Forest classifiers were trained to evaluate the predictive power of variables such as pebble shape, fissures, and rock type. Comparison between models based on original and synthetically balanced datasets revealed significant improvements in classification accuracy, particularly for unfractured samples. The results emphasize the central role of shape in fracture likelihood, and demonstrate the utility of synthetic resampling and causal inference for interpreting complex taphonomic processes in observational archaeological data. Experimental findings corroborate the observed patterns, supporting the potential generalizability of the proposed causal model across fire-affected contexts.