(PDF) Social scientists frequently make use of datasets where one or more variables are generated through human coding decisions (e.g. democracy scores, political party manifestos). 

But human-coded variables typically carry some amount of measurement error.