Here is my one page proposal for a joint geography and transport PhD on road traffic accident incidence submitted on 23/02/00:

Explaining Geographical Road Traffic Accident Patterns

Analysing road traffic accident (RTA) patterns is of considerable practical importance.  RTAs account for approximately 300000 fatalities worldwide, 4000 in the UK and 100 in West Yorkshire annually, the number of injuries is around 50 to 100 times higher.  Perhaps the most worrying thing of all is that despite attempts to improve safety, all these figures are rising.  Spatially referenced data concerning RTAs has been collected and stored as a mandatory requirement in the UK for many years.  This data is compiled in a database or data structure called STATS19 which contains grid references (giving the approximate location of accidents to the nearest metre), variables describing the time of the accident, the people and vehicles involved, and the road and weather conditions.  Further information in the database includes a classification for the severity of the accident and a written description of the event.  From STATS19 different types of accidents can be selected and the density of these accidents can be compared with those from a model that tries to predict these accident frequencies or rates.  Such models have been developed based on road network characteristics and other geographical information but there is likely to be a geographical residual or observable difference between what is predicted by the model and what is observed or recorded in the incidence data.  The Geographical Analysis Machine (GAM) can be used to map this residual error and identify where model predictions differ greatly from the observed incidence rates over a given period of time.  A geographical residual is very likely to exist because of exogenous geographical factors that affect RTA incidence that are not included in the models.  What are these exogenous geographical factors and can they be measured or used to develop better models and road safety information?  What are the different types of geographical factors that are related to RTA incidence?  What geographical information can be used to predict RTA rates?  How much of the observed RTA incidence can be explained by geographical information?  How much are RTA rates affected by the geographical environment?  What additional geographical variables can improve transport models that predict RTA rates or frequency?  What geographical information can help target road safety campaigns?  These kinds of questions will be addressed in the PhD and an introduction and extensive literature review will ground the research.

GAM is described and available on-line at