How do we avoid imposing the present on the past when modelling spatial interactions?