A0869
Title: Statistical matching methods for company data using multinomial logit models
Authors: Isao Takabe - Rissho University (Japan) [presenting]
Abstract: Statistical matching methods aim to build useful data by combining different data sources. These techniques make it possible to create informative data without conducting any surveys or collecting additional data. In recent years, various types of data have become available; as a result, statistical matching methods have been employed in various fields(e.g., econometrics, marketing, etc.). A statistical matching methodology is proposed by employing a multinomial logit model using weighted distance between records of data. This method makes it possible to efficiently and effectively perform statistical matching even for company data lacking detailed text information such as names and addresses. In addition, weights can be reasonably estimated based on company data, whereas there have not been previous studies that estimate the weights of distances. The techniques are applied to commercial company data and official economic census microdata. The results showed that the method performs better than the nearest neighbor methods in previous studies in terms of true match rate.
