Wilkinson's Tests and SQL Packages

Abstract

Wilkinson's Tests are used to benchmark the accuracy of some statistical functions in six SQL packages: Apache Hive, Microsoft Access, Microsoft SQL Server, MySQL, Oracle 11g SQL, and SAP Hana. Using the best choice of data type, we find that different packages use different rounding schemes, two packages use unreliable algorithms to compute the sample variance, one package returns the population standard deviation when the sample standard deviation is called, and one package has an unstable algorithm for computing the correlation coefficient. Using the wrong data type all but guarantees inaccurate results.