This thesis addresses two related issues. The primary issue is methodological and concerns the development of models, instruments and processes for realist evaluation research (Pawson and Tilley, 1997) in small scale, community based settings. Evaluation methods, however, must evaluate or investigate something', and so the secondary issue is the question to which the methods were applied in this case. That is, why is it that some early intervention programs which seem to 'work' for families at moderate levels of disadvantage seem to make things worse - specifically, making children's social and cognitive development outcomes worse - for those at more extreme levels of disadvantage? This may be conceptualised as the content question' for the thesis. The thesis is avowedly applied in intent. It seeks to demonstrate that realist evaluation methods may be applied in community based settings, and how they may be developed for particular contexts and questions. It also seeks to provide models and instruments which have been tested in one context and which can be used, or adapted for use, in other programs. Finally, it seeks to contribute both a theoretical framework and practical findings in relation to the content' question, which may inform future investigations into the impacts of early years programs, or indeed their design.