Using machine-learning methods in meta-analyses: An empirical application on consumer acceptance of meat alternatives

Abstract

Meta-analyses are widely used in various academic fields, including applied economics. However, the high labor intensity involved in paper searching and small sample sizes remain two dominant limiting factors. We conducted a meta-analysis of studies on consumer preferences for plant-based and lab-grown meat alternatives using machine-learning techniques at both the data collection and the data analysis phases. We demonstrated that machine learning reduces the workload in the manual title-abstract screen phase by 69% accounting for 24% of total workload in data collection. We also found that machine learning improves out-of-sample of sample prediction accuracy by 48–78 percentage points when compared to econometric model. Notably, we showed that integrating machine learning can also improve the predictive performance of econometric methods, thereby improving their out-of-sample predictions. Our empirical findings further revealed that demand for meat alternatives is higher among younger consumers, especially when the products displayed benefit information.