Design and analysis of experiments of embedded in complex sample surveys

Randomized experiments embedded in probability samples typically find their applications in survey methodology to test the effect of alternative survey implementations on the outcomes of a sample survey. The value of empirical research into survey methods is strengthened as conclusions can be generalized to populations larger than the sample that is included in the experiment. This can be achieved by selecting experimental units randomly from a larger target population and naturally leads to randomized experiments embedded in probability samples. This thesis develops statistical methods for design and analysis of such large scale field experiments.
An important issue in the analysis of such experiments is to find the right mode of inference. If experiments are embedded in probability samples with the purpose to generalise conclusions to larger target populations, a design-based mode of inference might be more appropriate than the model-based approach traditionally used in the analysis of randomized experiments. This requires an analysis procedure that account for the complexity of the sample design used to select a random sample from a target population as well as the experimental design used to assign the sampling units to the different treatments of the experiment. The main purpose of this thesis is to develop a general design-based frame work for design and analysis of experiments embedded in probability samples.