The design and analysis of research in the social sciences involve many hundreds if not thousands of assumptions and decisions, from how many villages will eventually be in a program to the statistical procedure to adjust for missing data. Each decision affects the value of the resulting inferences. Many researchers would prefer to commit themselves to a set of valid decisions before seeing the outcomes of the study, the best practice recommended across disciplines. Yet today, researchers do not have tools to characterize their research designs in a complete and transparent way. Moreover, researchers would like to be able to assess the statistical properties of their designs before making that set of binding decisions. Unfortunately, existing tools only allow crude assessments of the value of a design that ignore many complexities.
We propose a solution to these twin problems of an inability to plan and to commit to a design and analysis plan before the study begins. Our solution is to treat an entire design as an object with a set of minimal components that need to be specified. We provide a statistical package, DeclareDesign, that allows users to create this object and use it to generate design diagnostics, summaries, and comparisons. This provides a facility for users to learn about the probative value of an experiment before it is implemented and provides a way to formally specify research strategies that can be committed to ex ante.