Many promising Phase II trials have been carried out in head injury however to date there has been no successful translation of the positive results from these explanatory trials into improved patient outcomes in Phase III trials. Many reasons have been hypothesised for this failure. Outcomes in head injury trials are usually measured using the five point Glasgow Outcome Scale. Traditionally the ordinality of this scale is disregarded and it is dichotomised into two groups, favourable and unfavourable outcome. This thesis explores whether suboptimal statistical analysis techniques, including the dichotomisation of outcomes could have contributed to the reasons why Phase III trials have been unsuccessful. Based on eleven completed head injury studies, simulation modelling is used to compare outcome as assessed by the conventional dichotomy with both modelling that takes into account the ordered nature of the outcome (proportional odds modelling) and modelling which individualises a patient’s risk of a good or poor outcome ( the ‘sliding dichotomy’). The results of this modelling show that both analyses which use the full outcome scale and those which individualise risk show great efficiency gains (as measured by reduction in required sample sizes) over the conventional analysis of the binary outcome. These results are consistent both when the simulated treatment effects followed a proportional odds model and when they did not. Consistent results were also observed when targeting or restricting improvement to groups of subjects based on clinical characteristics or prognosis. Although proportional odds modelling shows consistently greater sample size reductions the choice of whether to use proportional odds modelling or the sliding dichotomy depends on the question of interest.