Statistical discrimination, Halo Effect, Cognitive Fluency and Measurement Error – Analyzing Grading Bias Caused by Handwriting Quality in A Randomized Controlled Trial
*Names in bold indicate Presenter
This bias has yet been estimated because content quality naturally correlates with handwriting quality, and handwriting is hard to quantify. This study utilizes a special rule of handwriting quality score in Nantong prefecture’s 9th grade reading exam to quantify handwriting quality (in China). This study randomly creates several handwritten versions for each one of 800 representative essays to break the natural tie between handwriting and content quality. This study also employs instrumental variable method to correct measurement error on the right-side variable and avoids the attenuation bias. The estimated bias is about 0.4-0.5 stand deviation, which means 1 point in handwriting(0-5) causes 2.2-2.7 points bias in content score(0-60).
Additional experiment designs break the mechanism of this bias quantitatively into statistical discrimination and taste discrimination, and separate taste discrimination into two cognitive biases (halo effect and cognitive fluency effect).
To reduce discrimination and miss-select talents, this study suggests that schools should implement typing in exams as early as possible, and developing an optical-character-recognition grading system is urgent. This experiment is the first one with clean separation of statistical discrimination and taste discrimination, and the first one to quantitatively estimate halo effect.