Abstract

This session has four speakers and a discussant representing four countries, with two of the speakers being females. In this session we will present results on theoretical, methodological and applications of some recent developments relating to the interplay between two topics of interest: non-probability surveys and small area methods. Presenters will illustrate the use of small area methods in leveraging data from social media data, web-scraping and natural language processing other non-traditional sources of information, along with regular probability-sampling driven surveys. The use of empirical likelihood methods and Bayesian network approaches would be highlighted through applications using data from multiple countries. Multiple speakers will also discuss doubly robust inferential schemes for small area methods when data from non-probability samples are also included. One of the speakers will elaborate on the use of data from social media, news websites and other non-traditional sources to augment the data from surveys and censuses. The contextual examples will be from multiple countries, showing the intricacies of the natural language processing and computations involved just to coupling the information from the data with traditional surveys. Then they will highlight the use of such interesting auxiliary information in small area statistics. Methodological developments as well as real data examples will be presented. Another speaker will then discuss the use of purposive sampling and non-probability sampling in the context of agricultural surveys and agricultural statistics. Here, some of the challenges involved require doubly robust inferential schemes and risk assessment, and methodological developments as well as case studies would be presented. Another presenter will highlight the use of empirical likelihood-based techniques in the context of doubly robust estimation using non-probability samples. Traditional approaches for using empirical likelihood do not extend easily when the sampling strategy is not driven by a probabilistic mechanism, and theoretical discussions as well and real data studies will be demonstrated. Another speaker will discuss the use of Bayesian network-driven small area modeling methodology when data from both non-probability and probability survey samples are combined. Certain interesting robustness features of this Bayesian approach would be highlighted. Using both theoretical results and real data case studies, the speaker will discuss how networks can be very informative auxiliary information for small area studies. We expect that the methodological advances as well as the real data examples from all the speakers would be of interest to many participants, and they would find useful information to take away and apply in other statistical applications as well.