A1249
Title: Sample integration using the bootstrap method and machine learning algorithms
Authors: Jorge Luis Rueda Sanchez - University of Granada (Spain) [presenting]
Maria del Mar Rueda - Universidad de Granada (Spain)
Beatriz Cobo Rodriguez - University of Granada (Spain)
Ramon Ferri-Garcia - University of Granada (Spain)
Abstract: Currently, multiple data sources are available in daily life, and managing and processing such information is a complex task. This problem also occurs in official statistical institutions, where two main sources of information are available: probability samples and non-probability samples. The latter differ in that their sampling design is unknown or suboptimal, which compromises representativity and produces bias in estimates. However, this source of information has great potential because it is obtained quickly, cheaply, and simply, unlike probability samples. Estimates from these samples are generally obtained through specialized estimators that use predictive models, in which machine learning algorithms can be employed. In the context of sample integration, non-probability samples are used to improve the precision of deficient probability samples. Different estimators exist for estimating population parameters in this context; however, they have room for improvement as they depend on certain assumptions that are difficult to meet in practice. Furthermore, variance estimation for these estimators is a complex and under-explored topic, and the bootstrap method can be a suitable approach. This study explores the use of the bootstrap method and machine learning algorithms for sample integration to improve the precision of estimators and to estimate their variance.
