How to reduce the number of rating scale items without predictability loss? Article in Scientometrics (February 2017)