Improving the Adaptability of Experimental Replication in Public Administration:A Bayesian Experimental Design Framework
In recent years,public administration experiments and their replication have be-come an important way to improve the generalizability of theories.However,experimental replica-tion still faces a series of methodological issues that need to be addressed:how to effectively estab-lish the relevance and comparability of replication with the original study?How to design and ana-lyze replication experiments reasonably and allow for flexible adjustments?How to optimize the de-sign to reduce sample size and cost,and to improve efficiency and effectiveness?This paper proposes a design framework based on Bayesian Experiments to provide a more adaptive path for replication of public administration experiments.Unlike Randomized Controlled Trials(RCTs)based on Frequen-cy Statistics,Bayesian Experiments can,first,take background knowledge such as original research as priors and perform Bayesian updating based on data and likelihood functions.Second,the use of posterior probabilities instead of using P-values to test research hypotheses avoids problems such as P-hacking.In a sequential setting,subsequent designs can be quickly adjusted based on previous re-sults,while ensuring comparability of results across experimental arms.It is also possible to deter-mine the location,context,and sample for further experimental replication based on structural spec-ulation.Finally,Bayesian experiments can be used to transform the intervention maximization prob-lem into a"multi-armed bandit problem"in reinforcement learning by using algorithms such as as Thompson sampling to deterministically assign subjects,it could significantly reduce the sample size and experimental cost,and has a wide range of application prospects.
万方数据
维普
