Research On A Standard Variable Selection Technique For Open-source Sociology

Recently,more and more researchers from natural science began to rethink the replicability of some research in their field.However,there are also many subdisciplines in social science use data to reflect social phenomena.Therefore,social science researchers also need to introspect the replicability of their research and do some related replicability tests.The replicability test which is focusing on verification and extension named “replication study”.To complete replication studies,we need to establish a data sharing and code sharing mechanism,i.e.“open-source sociology”.Since most researches do not meet the condition of open source in variable selection,it is difficult to ensure the implementation of both verification and extension.For this reason,it is particularly important to adopt a standard variable selection procedure.Taking the study of returns to education in rural areas as an example and using BIC and RMSE as indicators,this paper proposes a Bayesian model average method based variable filtering technique,which can simplify models with better goodness of fit,fitting capacity and prediction accuracy than other models.Therefore,the Bayesian model average method shows promising performance and can contribute to standardized variable selection in quantitative research in open-source sociology.In the end,this paper also explore the result robustness under MAR and MNAR circumstances,which can provide some reference for follow-up studies about its applicability.