Files in this item

FilesDescriptionFormat

application/pdf

application/pdfXU-DISSERTATION-2018.pdf (12MB)Restricted Access
(no description provided)PDF

Description

Title:Essays in applied microeconomics
Author(s):Xu, Jianfeng
Director of Research:Bernhardt, Mark Daniel
Doctoral Committee Chair(s):Bernhardt, Mark Daniel
Doctoral Committee Member(s):Deltas, George; Osman, Adam; Mazumder, Bhash
Department / Program:Economics
Discipline:Economics
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Handwriting
grader bias
discrimination
measurement error
school starting age
cognitive skills
non-cognitive skills
attenuation bias
Abstract:The Grading of the content may be biased against poor handwriting, resulting in discrimination against students with poor handwriting. Chapter one is the first to estimate the grading bias and decompose it to uncover its sources. I find that the bias alters 1.9% of high school admission outcomes and contributes to 70% of the gender gap in 9th grade’s writing test. The challenge with identifying bias is that ability may be correlated with handwriting quality. To quantify handwriting quality, I conduct a field experiment in a prefecture of China utilizing special rubrics for handwriting quality. To break the intrinsic correlation between handwriting and content quality, I randomly create two handwritten versions for each of 800 essays. The estimated bias is about 0.44 of a standard deviation: 1 point in handwriting (0-5) results in 2.45 points bias in content scores (0-60). Further experiments break the mechanism of this bias into statistical discrimination and Becker’s taste-based bias. I find statistical discrimination is negligible. In addition to true taste-based bias, I propose two cognitive biases (halo effect and cognitive fluency effect) and suggest halo effect is the major source of bias. To correct the attenuation bias caused by measurement error, I develop a new way to improve the small sample performance of traditional instrumental variable estimators. The second chapter investigates the influence of school starting age on cognitive and non-cognitive skills. Previous literature found a positive impact of relatively older school starting age on students’ cognitive skills based on national representative data from a few OECD countries. This study uses new survey data collected from all 77 schools in a county in China to analyze the effects of school starting age on both cognitive and non-cognitive skills. Measures of non-cognitive skills include Big-Five personality traits, locus of control, intrinsic motivation of study, persistence, and leadership. The cutoff-date rule of school entry is not strictly implemented in China, and the standard instrument in previous literature (assigned school starting age) is weak. However, I find that in one 6-month birth interval each year, parents rarely manipulate entry age. Utilizing the exogenous variations of school starting age within these less-manipulated six-month intervals, I correct the downward bias caused by parents’ manipulated early starting and find that one year older increases students’ reading score by 0.2 standard deviations for both genders. Among non-cognitive skills, I find being one year older boosts male students’ conscientiousness by 0.3 standard deviations and strength of leadership (0–5) by 0.46, but the effects on female students are smaller and insignificant. This causal relation between starting age and non-cognitive skills provides a new channel to suggest that Angrist and Krueger (1991)’s IV estimations may underestimate the true return to schooling. The third chapter studies the new instrumental variable estimator proposed in the first chapter. I perform intensive Monte Carlo simulations to compare the finite sample properties of several estimators, including OLS on average measurements (AVGOLS), bias-corrected OLS, TSLS, JIVE, LIML and two new estimators I propose: “Symmetric Averaging Instrumental Variable Estimator” (SAIV) and “Symmetric GMM Instrumental Variable Estimator” (SGMM). I suggest using AVGOLS, SGMM, SAIV and TSLS. SAIV dominates bias-corrected OLS, TSLS, JIVE and LIML in central tendency and dispersion. I also investigate the number of measurements needed in experiment/survey design and find that SAIV requires a smaller number of measurements.
Issue Date:2018-12-07
Type:Thesis
URI:http://hdl.handle.net/2142/102936
Rights Information:Copyright 2018 Jianfeng Xu
Date Available in IDEALS:2019-02-08
Date Deposited:2018-12


This item appears in the following Collection(s)

Item Statistics