Past Exams
Exam 2022
Question 1
- Test many outcomes would cause Multiple Inference Problem
- 在
Autumn Topic 2曾经出现过,“In this context, if you test multiple hypotheses (e.g. effect of program on multiple outcomes) you might find some significant coefficients by chance” 课件介绍了两个解决方法: -
- Summary index approach — 把多个相关outcomes标准化后加权平均成一个index,然后只对这个index做一次test。这是Anderson (2008)提出的方法,slides详细列出了5个步骤。
-
- FWER(Familywise Error Rate Control) — 这就是调整significance level的方法,Bonferroni correction是其中最简单的一种。基本逻辑是:如果你测试 KK K 个hypotheses,就把significance level从 α\alpha α 调整为 α/K\alpha / K α/K。比如测试20个outcomes,每个test用 0.05/20=0.00250.05/20 = 0.0025 0.05/20=0.0025 作为threshold。
- 在
Exam 2024
Question 2
- 为什么用弹性而不是导数?
- 因为在比较两个对象的时候往往量纲不一样。弹性是一种无量纲的百分比对百分比的相应度,因此方便进行跨国家,跨时间的比较。
- d 问,值得 criticize 的点:相对工资的下降并不等于绝对工资的下降。例如在题目设定的条件下, 持续上升,如果 ,也就是说两者是 gross substitute 的话,
Question 3
- Becker Model