To improve the provision of health services, policymakers require predictions of the effectiveness and cost-effectiveness of future policy changes. A common way of predicting the impact of future policy change is by using responses from stated preference surveys, to predict the response of the target population of interest. However, this approach assumes that stated preference models accurately predict actual behaviour (revealed preference) in the population of interest. The increasing availability of large observational datasets can allow this assumption to be carefully assessed. These observational data can then inform the design and analysis of stated preference surveys, to minimise hypothetical bias – the discrepancy between stated and revealed preferences. This thesis developed and applied an approach for eliciting stated preferences to improve predictions of the effects of future policies in the context of alternative strategies for improving blood collection services in England. The aim of investigating the external validity of stated preferences has two underlying objectives. The first objective is to develop an approach for reducing hypothetical bias in the prediction of revealed preferences from responses to stated preference surveys. The second objective is to compare estimates of relative preferences from stated preference questions that require ordered categorical versus discrete choice responses. The thesis finds that an approach which combines an iterative design (ex ante) and a mediation approach which explicitly recognises constraints (ex post) reduces estimates of hypothetical bias. It also finds that a survey design that requires ordered categorical versus discrete choice responses provide similar estimates of marginal rates of substitution. The stated preference tasks that require ordered categorical responses have the advantage that they can provide estimates (in this case of blood donation frequency) that are of direct use for policymaking. The thesis concludes that the approach developed to eliciting stated preferences can reduce hypothetical bias and provide estimates of direct policy-making relevance.