Exploring the use of a lasso algorithm in a moderator problem in meta-analysis

Abstract

In meta-regression the amount of measured of measured moderators is often fairly high relative to the amount of studies included in the sample. To solve this regression problem, conventional method such as the Ordinary Least Squares method lack the ability the predict future data under these circumstances. To solve this problem, the OLS is substituted for a LASSO regression, which decreases the variance in future data by increasing the bias. This study provides some evidence that meta-regression with LASSO is a solution for the situation where the amount of measured moderators is high, and the amount of studies included in the sample is low. However, the use of LASSO regression in meta-analysis should be explored further.