Since the 1990s, the use of movement models for predicting and explaining past land routes has become common-place in archaeology. Despite their long use, movement models remain largely theoretically unjustified and methodologically immature. Without addressing these issues, outcomes from movement models will continue to lack a theoretical connection to the land route being examined, and thus remain unreliable, offering predictions and explanations that are epistemologically weak.

This thesis investigates the generative processes behind land routes created in the past, using Roman roads as its case study. Theoretically informed and methodologically robust movement models are developed for examining the factors that influenced the placement of Roman roads in Britain and Sardinia, how these factors differed across roads, and what these factors reveal about the structure of individual Roman provinces and the Roman Empire.

First, the methodological robustness of outcomes from movement models was assessed in light of error in digital elevation models used to represent the landscape. This was tackled through the use of a novel Monte-Carlo simulation approach, whereby error in the digital elevation model is propagated throughout the analysis to produce a probabilistic outcome. This is demonstrated through the explanation of a Roman road in north-west Britain, finding that minimising energy expenditure can only explain sections of the Roman road.

Second, the impact of how movement models formally represent hypotheses was examined using Multiple Model Idealisation and Robustness Analysis. Through the use of computer simulations under known conditions, it is shown that whilst model outcomes from movement models representing the same hypothesis produce similar results, model outcomes representing different hypotheses can also produce similar results. Thus, hypotheses as represented by different movement models have the risk of being underdetermined when used for explaining land routes created in the past. These insights were then applied for the explanation of a Roman road in Sardinia, Italy, showing that the placement of the road was not constructed to minimise time taken or energy expended.

Finally, theoretically informed and methodologically robust movement models were developed for the explanation of Roman roads in the British section of the Antonine Itinerary. Through the use of Approximate Bayesian Computation and Multi-Criteria Decision Analysis, it is shown that the influence of factors on the placement of Roman roads was not homogenous but rather reflects the specific circumstances that were present during their individual construction.