Using Machine Learning to Evaluate Questions in a Multilingual Survey

Machine learning is becoming an increasingly popular tool in survey data collection. We recently conducted a successful application of machine learning to automatically detect survey questions at a higher risk of poor performance. The proof-of-concept study utilized items from an establishment survey and demonstrated that the machine learning methodology was able to identify the worst-performing items as evaluated by survey methodology experts. This paper extends ongoing research by applying the methodology to assess the performance of items in a multilingual household survey. In particular, we compared the performance of survey questions in both English and Spanish interviews. The results show that the use of machine learning identified a common set of items as difficult and prone to poor performance in both English and Spanish interviews. The implications of the findings are discussed.