﻿[Script Info] Title: [Events] Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text Dialogue: 0,0:00:00.30,0:00:04.64,Default,,0000,0000,0000,,因果推断之路径既黑暗又危险 Dialogue: 0,0:00:05.14,0:00:08.02,Default,,0000,0000,0000,,但是计量经济学是很厉害的武器 Dialogue: 0,0:00:08.48,0:00:11.73,Default,,0000,0000,0000,,当自然界给你带来偶然的随机分配时 Dialogue: 0,0:00:11.73,0:00:15.80,Default,,0000,0000,0000,,使用气势汹汹与灵活多變的\N工具变量进行攻击 Dialogue: 0,0:00:19.39,0:00:21.09,Default,,0000,0000,0000,,[] Dialogue: 0,0:00:23.65,0:00:26.36,Default,,0000,0000,0000,,随机试验是完成\N“其他条件不变”的比较 Dialogue: 0,0:00:26.36,0:00:28.70,Default,,0000,0000,0000,,的最可靠途径 Dialogue: 0,0:00:28.70,0:00:32.64,Default,,0000,0000,0000,,但我们经常无法使用\N这个功能强大的工具 Dialogue: 0,0:00:33.22,0:00:36.94,Default,,0000,0000,0000,,但是有时候，随机是偶然发生的 Dialogue: 0,0:00:36.94,0:00:40.59,Default,,0000,0000,0000,,这时候我们转向工具变量 Dialogue: 0,0:00:40.59,0:00:41.94,Default,,0000,0000,0000,,—简称IV Dialogue: 0,0:00:41.94,0:00:44.51,Default,,0000,0000,0000,,工具变量 Dialogue: 0,0:00:44.51,0:00:48.19,Default,,0000,0000,0000,,今天的课堂是IV两节课的第一节 Dialogue: 0,0:00:48.96,0:00:52.80,Default,,0000,0000,0000,,我们的第一节IV课\N从学校的故事开始 Dialogue: 0,0:00:52.80,0:00:54.35,Default,,0000,0000,0000,,[] Dialogue: 0,0:00:54.35,0:00:56.14,Default,,0000,0000,0000,,特许学校是一些公立学校 Dialogue: 0,0:00:56.14,0:01:00.11,Default,,0000,0000,0000,,不受日常学区监督\N与教师工会合同约束 Dialogue: 0,0:01:00.90,0:01:03.51,Default,,0000,0000,0000,,特许学校能否提高成绩 Dialogue: 0,0:01:03.51,0:01:05.16,Default,,0000,0000,0000,,是美国教育改革史上 Dialogue: 0,0:01:05.16,0:01:07.76,Default,,0000,0000,0000,,最重要的问题之一 Dialogue: 0,0:01:08.14,0:01:12.56,Default,,0000,0000,0000,,最受欢迎的特许学校的申请人数\N远多于学位 Dialogue: 0,0:01:12.56,0:01:16.46,Default,,0000,0000,0000,,因此抽奖运决定了\N谁家孩子可获录取 Dialogue: 0,0:01:16.87,0:01:20.70,Default,,0000,0000,0000,,在学生争夺机会时需要面对很多风险 Dialogue: 0,0:01:20.70,0:01:25.00,Default,,0000,0000,0000,,正如获奖纪录片“等待超人”中 Dialogue: 0,0:01:25.00,0:01:27.83,Default,,0000,0000,0000,,所描述的那样 Dialogue: 0,0:01:27.83,0:01:29.70,Default,,0000,0000,0000,,等待结果时会产生很多种情绪 Dialogue: 0,0:01:30.26,0:01:32.92,Default,,0000,0000,0000,,别哭，你会让妈妈哭的\N好吗？ Dialogue: 0,0:01:37.50,0:01:40.62,Default,,0000,0000,0000,,特许学校真的能提供更好的教育吗？ Dialogue: 0,0:01:40.95,0:01:43.18,Default,,0000,0000,0000,,评论家肯定会说"不是的" Dialogue: 0,0:01:43.41,0:01:46.59,Default,,0000,0000,0000,,他们会争辩说特许学校\N能夠招募更好 Dialogue: 0,0:01:46.59,0:01:50.16,Default,,0000,0000,0000,,更聪明或更主动的学生\N因此以后结果的差异 Dialogue: 0,0:01:50.16,0:01:52.06,Default,,0000,0000,0000,,反映了选择性偏差 Dialogue: 0,0:01:52.60,0:01:54.73,Default,,0000,0000,0000,,等一下，这个似乎很容易 Dialogue: 0,0:01:55.14,0:01:57.64,Default,,0000,0000,0000,,在抽奖活动中\N我们会随机选择优胜者 Dialogue: 0,0:01:57.64,0:02:00.08,Default,,0000,0000,0000,,因此只比较赢家和输家\N很明显的 Dialogue: 0,0:02:00.08,0:02:01.78,Default,,0000,0000,0000,,在正确的轨道上，卡马尔 Dialogue: 0,0:02:01.78,0:02:04.38,Default,,0000,0000,0000,,但是特许学校的抽签安排 Dialogue: 0,0:02:04.38,0:02:07.56,Default,,0000,0000,0000,,不会强迫孩子们进入\N或离开特定的学校 Dialogue: 0,0:02:07.75,0:02:10.67,Default,,0000,0000,0000,,他们随机分配了特许学校的学位 Dialogue: 0,0:02:11.65,0:02:13.45,Default,,0000,0000,0000,,有些孩子很幸运 Dialogue: 0,0:02:13.45,0:02:14.97,Default,,0000,0000,0000,,有些孩子不是 Dialogue: 0,0:02:14.97,0:02:17.24,Default,,0000,0000,0000,,如果我们只是想知道特许学校 Dialogue: 0,0:02:17.24,0:02:19.20,Default,,0000,0000,0000,,所带来的影响 Dialogue: 0,0:02:19.20,0:02:22.42,Default,,0000,0000,0000,,我们可以将其视为随机试验 Dialogue: 0,0:02:22.72,0:02:24.68,Default,,0000,0000,0000,,但是，我们只对特许学校\N就学的影响 Dialogue: 0,0:02:24.68,0:02:27.04,Default,,0000,0000,0000,,感兴趣 Dialogue: 0,0:02:27.04,0:02:28.28,Default,,0000,0000,0000,,而对录取不感兴趣 Dialogue: 0,0:02:28.57,0:02:32.04,Default,,0000,0000,0000,,并非所有获录取的学生\N都会接受学位 Dialogue: 0,0:02:32.04,0:02:37.23,Default,,0000,0000,0000,,IV将被录取为特许学校学生的影响 Dialogue: 0,0:02:37.23,0:02:40.37,Default,,0000,0000,0000,,转变为实际就读特许学校的影响 Dialogue: 0,0:02:40.37,0:02:42.34,Default,,0000,0000,0000,,- 太酷了\N- 哦，太好了 Dialogue: 0,0:02:45.92,0:02:48.87,Default,,0000,0000,0000,,让我们看一个例子 Dialogue: 0,0:02:48.87,0:02:52.35,Default,,0000,0000,0000,,这是一所执行知识就是力量专案\N的特许学校，或简称为KIPP Dialogue: 0,0:02:52.74,0:02:54.94,Default,,0000,0000,0000,,这所KIPP特许学校位于林恩 Dialogue: 0,0:02:54.94,0:02:58.84,Default,,0000,0000,0000,,一座位于麻省海边的\N褪色工业城镇 Dialogue: 0,0:02:59.10,0:03:01.89,Default,,0000,0000,0000,,这所学校的申请者多于学位 Dialogue: 0,0:03:01.89,0:03:05.62,Default,,0000,0000,0000,,因此他们要抽签来挑选学生 Dialogue: 0,0:03:05.83,0:03:11.85,Default,,0000,0000,0000,,从2005年到2008年\N共有371名四年级以及五年级生 Dialogue: 0,0:03:11.85,0:03:15.35,Default,,0000,0000,0000,,参加了KIPP林恩的抽签 Dialogue: 0,0:03:15.35,0:03:18.80,Default,,0000,0000,0000,,当中253名学生KIPP获录取 Dialogue: 0,0:03:18.80,0:03:21.65,Default,,0000,0000,0000,,118名学生没有录取 Dialogue: 0,0:03:21.97,0:03:26.00,Default,,0000,0000,0000,,一年后，获录取者的数学分数 Dialogue: 0,0:03:26.00,0:03:27.85,Default,,0000,0000,0000,,比未获录取者更高 Dialogue: 0,0:03:27.85,0:03:30.47,Default,,0000,0000,0000,,我们并不是试图弄清楚 Dialogue: 0,0:03:30.47,0:03:33.80,Default,,0000,0000,0000,,获录取后是否会提高\N你的数学水平 Dialogue: 0,0:03:34.07,0:03:38.47,Default,,0000,0000,0000,,我们想知道参加KIPP\N是否会使你的数学成绩改进 Dialogue: 0,0:03:39.04,0:03:45.75,Default,,0000,0000,0000,,在253位获录取者中\N实际上只有199位到KIPP上学 Dialogue: 0,0:03:46.14,0:03:48.80,Default,,0000,0000,0000,,其他学生选择了传统的公立学校 Dialogue: 0,0:03:49.56,0:03:55.54,Default,,0000,0000,0000,,同样，在118名未被录取的学生中\N事实上有一些最终参加了KIPP Dialogue: 0,0:03:55.54,0:03:57.45,Default,,0000,0000,0000,,他们后来也获录取 Dialogue: 0,0:03:57.45,0:04:00.04,Default,,0000,0000,0000,,那么，实际上参加KIPP Dialogue: 0,0:04:00.04,0:04:02.38,Default,,0000,0000,0000,,对考试成绩有何影响呢？ Dialogue: 0,0:04:03.11,0:04:05.43,Default,,0000,0000,0000,,为什么我们不能只衡量\N他们的数学成绩？ Dialogue: 0,0:04:05.89,0:04:07.24,Default,,0000,0000,0000,,这是很好的问题 Dialogue: 0,0:04:07.24,0:04:09.30,Default,,0000,0000,0000,,你将他们与谁进行比较呢？ Dialogue: 0,0:04:09.30,0:04:11.11,Default,,0000,0000,0000,,那些没有参加的学生 Dialogue: 0,0:04:11.11,0:04:12.94,Default,,0000,0000,0000,,上学率是随机的吗？ Dialogue: 0,0:04:14.16,0:04:16.18,Default,,0000,0000,0000,,- 不是啊\N- 选择性偏差 Dialogue: 0,0:04:16.18,0:04:17.91,Default,,0000,0000,0000,,- 对啊\N- 什么？ Dialogue: 0,0:04:17.91,0:04:21.83,Default,,0000,0000,0000,,KIPP的录取是随机的，因此我们\N对“其他条件不变”的假设充满信心 Dialogue: 0,0:04:21.83,0:04:26.41,Default,,0000,0000,0000,,但上学率不是随机的 Dialogue: 0,0:04:26.64,0:04:30.63,Default,,0000,0000,0000,,选择接受录取通知 Dialogue: 0,0:04:30.63,0:04:32.98,Default,,0000,0000,0000,,可能是与数学成绩有关的特征 Dialogue: 0,0:04:33.25,0:04:36.16,Default,,0000,0000,0000,,例如，有奉献精神的父母 Dialogue: 0,0:04:36.16,0:04:38.96,Default,,0000,0000,0000,,更有可能接受录取 Dialogue: 0,0:04:38.96,0:04:42.65,Default,,0000,0000,0000,,无论上那间学校 Dialogue: 0,0:04:42.65,0:04:44.09,Default,,0000,0000,0000,,他们的孩子的数学成绩\N也有可能更好 Dialogue: 0,0:04:44.09,0:04:45.11,Default,,0000,0000,0000,,对啊 Dialogue: 0,0:04:45.11,0:04:47.72,Default,,0000,0000,0000,,IV将录取的影响 Dialogue: 0,0:04:47.72,0:04:50.57,Default,,0000,0000,0000,,转化为KIPP上学率的影响 Dialogue: 0,0:04:50.57,0:04:53.37,Default,,0000,0000,0000,,并就一些获录取者到其他学校上学 Dialogue: 0,0:04:53.37,0:04:56.57,Default,,0000,0000,0000,,而一些未被录取者还是设法\N参加了KIPP 而进行调整 Dialogue: 0,0:04:56.95,0:05:00.52,Default,,0000,0000,0000,,本质上，IV需要进行不完全的随机化 Dialogue: 0,0:05:00.52,0:05:03.01,Default,,0000,0000,0000,,并进行适当的调整 Dialogue: 0,0:05:03.68,0:05:07.11,Default,,0000,0000,0000,,怎么样？ IV描述了一种连锁反应 Dialogue: 0,0:05:07.64,0:05:10.34,Default,,0000,0000,0000,,为什么学校的录取会影响成绩？ Dialogue: 0,0:05:10.34,0:05:13.26,Default,,0000,0000,0000,,可能是因为这影响了\N特许学校的上学率 Dialogue: 0,0:05:13.26,0:05:16.64,Default,,0000,0000,0000,,而特许学校的上学率\N提高了数学成绩 Dialogue: 0,0:05:16.64,0:05:20.64,Default,,0000,0000,0000,,连锁反应的第一个环节\N称之为“第一阶段” Dialogue: 0,0:05:20.64,0:05:24.48,Default,,0000,0000,0000,,是抽签对特许学校上学率的影响 Dialogue: 0,0:05:24.48,0:05:28.45,Default,,0000,0000,0000,,第二阶段是在特许学校学 Dialogue: 0,0:05:28.45,0:05:30.15,Default,,0000,0000,0000,,以及结果变量之间的关联 Dialogue: 0,0:05:30.15,0:05:32.26,Default,,0000,0000,0000,,在这情况下，数学分数 Dialogue: 0,0:05:32.73,0:05:36.44,Default,,0000,0000,0000,,工具变量或简称为“工具” Dialogue: 0,0:05:36.44,0:05:40.25,Default,,0000,0000,0000,,是启动链式反应的变量 Dialogue: 0,0:05:40.98,0:05:43.99,Default,,0000,0000,0000,,工具变量对结果的影响 Dialogue: 0,0:05:43.99,0:05:46.63,Default,,0000,0000,0000,,称为简化式 Dialogue: 0,0:05:48.14,0:05:51.87,Default,,0000,0000,0000,,这个链式反应可以用数学表示 Dialogue: 0,0:05:51.87,0:05:54.24,Default,,0000,0000,0000,,我们乘以第一阶段 Dialogue: 0,0:05:54.24,0:05:56.35,Default,,0000,0000,0000,,即录取者对上学率的影响 Dialogue: 0,0:05:56.35,0:05:57.96,Default,,0000,0000,0000,,到第二阶段 Dialogue: 0,0:05:57.96,0:06:00.54,Default,,0000,0000,0000,,上学率对分数的影响 Dialogue: 0,0:06:00.54,0:06:02.71,Default,,0000,0000,0000,,我们得到简化式 Dialogue: 0,0:06:02.71,0:06:05.68,Default,,0000,0000,0000,,获录取对分数的影响 Dialogue: 0,0:06:06.78,0:06:11.57,Default,,0000,0000,0000,,简化式和第一阶段是可观察的\N并且易于计算 Dialogue: 0,0:06:11.75,0:06:14.88,Default,,0000,0000,0000,,但是，上学率对成绩的影响 Dialogue: 0,0:06:14.88,0:06:17.09,Default,,0000,0000,0000,,并未能直接观察到 Dialogue: 0,0:06:17.09,0:06:20.36,Default,,0000,0000,0000,,这是我们试图确定的因果关系 Dialogue: 0,0:06:21.04,0:06:23.83,Default,,0000,0000,0000,,Given some important assumptions\Nwe'll discuss shortly, Dialogue: 0,0:06:23.83,0:06:25.98,Default,,0000,0000,0000,,we can find the effect\Nof KIPP attendance Dialogue: 0,0:06:25.98,0:06:29.26,Default,,0000,0000,0000,,by dividing the reduced form\Nby the first stage. Dialogue: 0,0:06:29.26,0:06:32.91,Default,,0000,0000,0000,,This will become more clear\Nas we work through an example. Dialogue: 0,0:06:32.91,0:06:34.21,Default,,0000,0000,0000,,- [Student] Let's do this. Dialogue: 0,0:06:37.16,0:06:38.73,Default,,0000,0000,0000,,- A quick note on measurement. Dialogue: 0,0:06:38.73,0:06:41.74,Default,,0000,0000,0000,,We measure achievement\Nusing standard deviations, Dialogue: 0,0:06:41.74,0:06:44.73,Default,,0000,0000,0000,,often denoted\Nby the Greek letter sigma (σ). Dialogue: 0,0:06:44.73,0:06:48.86,Default,,0000,0000,0000,,One σ is a huge move\Nfrom around the bottom 15% Dialogue: 0,0:06:48.86,0:06:51.63,Default,,0000,0000,0000,,to the middle of most\Nachievement distributions. Dialogue: 0,0:06:51.63,0:06:55.41,Default,,0000,0000,0000,,Even a ¼ or ½ σ difference is big. Dialogue: 0,0:06:56.26,0:06:58.39,Default,,0000,0000,0000,,- [Instructor] Now we're ready\Nto plug some numbers Dialogue: 0,0:06:58.39,0:07:01.66,Default,,0000,0000,0000,,into the equation\Nwe introduced earlier. Dialogue: 0,0:07:01.66,0:07:03.23,Default,,0000,0000,0000,,First up, what's the effect Dialogue: 0,0:07:03.23,0:07:06.08,Default,,0000,0000,0000,,of winning the lottery\Non math scores? Dialogue: 0,0:07:06.35,0:07:10.42,Default,,0000,0000,0000,,KIPP applicants' math scores\Nare a third of a standard deviation Dialogue: 0,0:07:10.42,0:07:11.84,Default,,0000,0000,0000,,below the state average Dialogue: 0,0:07:11.84,0:07:14.39,Default,,0000,0000,0000,,in the year before\Nthey apply to KIPP. Dialogue: 0,0:07:14.39,0:07:18.32,Default,,0000,0000,0000,,But a year later, lottery winners\Nscore right at the state average, Dialogue: 0,0:07:18.32,0:07:21.48,Default,,0000,0000,0000,,while the lottery losers\Nare still well behind Dialogue: 0,0:07:21.48,0:07:25.50,Default,,0000,0000,0000,,with an average score\Naround -0.36 σ. Dialogue: 0,0:07:25.83,0:07:29.62,Default,,0000,0000,0000,,The effect of winning the lottery\Non scores is the difference Dialogue: 0,0:07:29.62,0:07:32.82,Default,,0000,0000,0000,,between the winners' scores\Nand the losers' scores. Dialogue: 0,0:07:33.40,0:07:35.78,Default,,0000,0000,0000,,Take the winners'\Naverage math scores, Dialogue: 0,0:07:35.78,0:07:38.27,Default,,0000,0000,0000,,subtract the losers'\Naverage math scores, Dialogue: 0,0:07:38.27,0:07:41.50,Default,,0000,0000,0000,,and you will have 0.36 σ. Dialogue: 0,0:07:41.91,0:07:46.88,Default,,0000,0000,0000,,Next up: what's the effect\Nof winning the lottery on attendance? Dialogue: 0,0:07:46.88,0:07:49.19,Default,,0000,0000,0000,,In other words,\Nif you win the lottery, Dialogue: 0,0:07:49.19,0:07:52.26,Default,,0000,0000,0000,,how much more likely\Nare you to attend KIPP Dialogue: 0,0:07:52.26,0:07:53.46,Default,,0000,0000,0000,,than if you lose? Dialogue: 0,0:07:53.67,0:07:57.80,Default,,0000,0000,0000,,First, what percentage\Nof lottery winners attend KIPP? Dialogue: 0,0:07:57.80,0:08:00.77,Default,,0000,0000,0000,,Divide the number of winners\Nwho attended KIPP Dialogue: 0,0:08:00.77,0:08:05.49,Default,,0000,0000,0000,,by the total number\Nof lottery winners -- that's 78%. Dialogue: 0,0:08:05.81,0:08:09.33,Default,,0000,0000,0000,,To find the percentage\Nof lottery losers who attended KIPP, Dialogue: 0,0:08:09.33,0:08:12.33,Default,,0000,0000,0000,,we divide the number of losers\Nwho attended KIPP Dialogue: 0,0:08:12.33,0:08:16.86,Default,,0000,0000,0000,,by the total number\Nof lottery losers -- that's 4%. Dialogue: 0,0:08:17.38,0:08:21.60,Default,,0000,0000,0000,,Subtract 4 from 78, and we find\Nthat winning the lottery Dialogue: 0,0:08:21.60,0:08:25.60,Default,,0000,0000,0000,,makes you 74%\Nmore likely to attend KIPP. Dialogue: 0,0:08:25.95,0:08:28.53,Default,,0000,0000,0000,,Now we can find\Nwhat we're really after -- Dialogue: 0,0:08:28.53,0:08:34.55,Default,,0000,0000,0000,,the effect of attendance on scores,\Nby dividing 0.36 by 0.74. Dialogue: 0,0:08:34.79,0:08:37.58,Default,,0000,0000,0000,,Attending KIPP raises math scores Dialogue: 0,0:08:37.58,0:08:41.61,Default,,0000,0000,0000,,by 0.48 standard deviations\Non average. Dialogue: 0,0:08:42.13,0:08:44.50,Default,,0000,0000,0000,,That's an awesome achievement gain, Dialogue: 0,0:08:44.50,0:08:47.38,Default,,0000,0000,0000,,equal to moving\Nfrom about the bottom third Dialogue: 0,0:08:47.38,0:08:49.92,Default,,0000,0000,0000,,to the middle\Nof the achievement distribution. Dialogue: 0,0:08:49.92,0:08:51.08,Default,,0000,0000,0000,,- [Student] Whoa, half a sig. Dialogue: 0,0:08:51.08,0:08:53.51,Default,,0000,0000,0000,,- [Instructor] These estimates\Nare for kids opting in Dialogue: 0,0:08:53.51,0:08:54.78,Default,,0000,0000,0000,,to the KIPP lottery, Dialogue: 0,0:08:54.78,0:08:57.76,Default,,0000,0000,0000,,whose enrollment status\Nis changed by winning. Dialogue: 0,0:08:57.98,0:09:00.62,Default,,0000,0000,0000,,That's not necessarily\Na random sample Dialogue: 0,0:09:00.62,0:09:02.28,Default,,0000,0000,0000,,of all children in Lynn. Dialogue: 0,0:09:02.54,0:09:05.04,Default,,0000,0000,0000,,So we can't assume\Nwe'd see the same effect Dialogue: 0,0:09:05.04,0:09:07.33,Default,,0000,0000,0000,,for other types of students.\N- [Student] Huh. Dialogue: 0,0:09:07.33,0:09:10.22,Default,,0000,0000,0000,,- But this effect\Non keen for KIPP kids Dialogue: 0,0:09:10.22,0:09:13.37,Default,,0000,0000,0000,,is likely to be a good indicator\Nof the consequences Dialogue: 0,0:09:13.37,0:09:15.77,Default,,0000,0000,0000,,of adding additional charter seats. Dialogue: 0,0:09:15.77,0:09:17.22,Default,,0000,0000,0000,,- [Student] Cool.\N- [Student] Got it. Dialogue: 0,0:09:19.63,0:09:23.35,Default,,0000,0000,0000,,- IV eliminates selection bias,\Nbut like all of our tools, Dialogue: 0,0:09:23.35,0:09:25.62,Default,,0000,0000,0000,,the solution builds on a set\Nof assumptions Dialogue: 0,0:09:25.62,0:09:27.54,Default,,0000,0000,0000,,not to be taken for granted. Dialogue: 0,0:09:28.10,0:09:31.46,Default,,0000,0000,0000,,First, there must be\Na substantial first stage -- Dialogue: 0,0:09:31.46,0:09:35.56,Default,,0000,0000,0000,,that is the instrumental variable,\Nwinning or losing the lottery, Dialogue: 0,0:09:35.56,0:09:39.06,Default,,0000,0000,0000,,must really change the variable\Nwhose effect we're interested in -- Dialogue: 0,0:09:39.06,0:09:41.03,Default,,0000,0000,0000,,here, KIPP attendance. Dialogue: 0,0:09:41.30,0:09:44.59,Default,,0000,0000,0000,,In this case, the first stage\Nis not really in doubt. Dialogue: 0,0:09:44.59,0:09:47.89,Default,,0000,0000,0000,,Winning the lottery makes\NKIPP attendance much more likely. Dialogue: 0,0:09:48.39,0:09:50.63,Default,,0000,0000,0000,,Not all IV stories are like that. Dialogue: 0,0:09:51.32,0:09:53.70,Default,,0000,0000,0000,,Second, the instrument\Nmust be as good Dialogue: 0,0:09:53.70,0:09:54.93,Default,,0000,0000,0000,,as randomly assigned, Dialogue: 0,0:09:54.93,0:09:58.72,Default,,0000,0000,0000,,meaning lottery winners and losers\Nhave similar characteristics. Dialogue: 0,0:09:58.89,0:10:01.56,Default,,0000,0000,0000,,This is the independence assumption. Dialogue: 0,0:10:01.98,0:10:05.72,Default,,0000,0000,0000,,Of course, KIPP lottery wins\Nreally are randomly assigned. Dialogue: 0,0:10:05.72,0:10:09.66,Default,,0000,0000,0000,,Still, we should check for balance\Nand confirm that winners and losers Dialogue: 0,0:10:09.66,0:10:11.49,Default,,0000,0000,0000,,have similar family backgrounds, Dialogue: 0,0:10:11.49,0:10:13.59,Default,,0000,0000,0000,,similar aptitudes and so on. Dialogue: 0,0:10:13.59,0:10:16.97,Default,,0000,0000,0000,,In essence, we're checking\Nto ensure KIPP lotteries are fair Dialogue: 0,0:10:16.97,0:10:20.06,Default,,0000,0000,0000,,with no group of applicants\Nsuspiciously likely to win. Dialogue: 0,0:10:21.37,0:10:24.37,Default,,0000,0000,0000,,Finally, we require\Nthe instrument change outcomes Dialogue: 0,0:10:24.37,0:10:26.09,Default,,0000,0000,0000,,solely through\Nthe variable of interest -- Dialogue: 0,0:10:26.09,0:10:28.10,Default,,0000,0000,0000,,in this case, attending KIPP. Dialogue: 0,0:10:28.30,0:10:31.37,Default,,0000,0000,0000,,This assumption is called\Nthe exclusion restriction. Dialogue: 0,0:10:32.95,0:10:37.50,Default,,0000,0000,0000,,- IV only works if you can satisfy\Nthese three assumptions. Dialogue: 0,0:10:38.03,0:10:40.42,Default,,0000,0000,0000,,- I don't understand\Nthe exclusion restriction. Dialogue: 0,0:10:40.92,0:10:43.60,Default,,0000,0000,0000,,How could winning the lottery\Naffect math scores Dialogue: 0,0:10:43.60,0:10:45.24,Default,,0000,0000,0000,,other than by attending KIPP? Dialogue: 0,0:10:45.24,0:10:47.23,Default,,0000,0000,0000,,- [Student] Yeah.\N- [Instructor] Great question. Dialogue: 0,0:10:47.23,0:10:50.54,Default,,0000,0000,0000,,Suppose lottery winners\Nare just thrilled to win, Dialogue: 0,0:10:50.54,0:10:55.04,Default,,0000,0000,0000,,and this happiness motivates them\Nto study more and learn more math, Dialogue: 0,0:10:55.04,0:10:57.28,Default,,0000,0000,0000,,regardless of where\Nthey go to school. Dialogue: 0,0:10:57.28,0:10:59.90,Default,,0000,0000,0000,,This would violate\Nthe exclusion restriction Dialogue: 0,0:10:59.90,0:11:03.79,Default,,0000,0000,0000,,because the motivational effect\Nof winning is a second channel Dialogue: 0,0:11:03.79,0:11:06.57,Default,,0000,0000,0000,,whereby lotteries\Nmight affect test scores. Dialogue: 0,0:11:06.86,0:11:09.55,Default,,0000,0000,0000,,While it's hard\Nto rule this out entirely, Dialogue: 0,0:11:09.55,0:11:12.65,Default,,0000,0000,0000,,there's no evidence\Nof any alternative channels Dialogue: 0,0:11:12.65,0:11:14.11,Default,,0000,0000,0000,,in the KIPP study. Dialogue: 0,0:11:17.82,0:11:20.70,Default,,0000,0000,0000,,- IV solves the problem\Nof selection bias Dialogue: 0,0:11:20.70,0:11:25.05,Default,,0000,0000,0000,,in scenarios like the KIPP lottery\Nwhere treatment offers are random Dialogue: 0,0:11:25.05,0:11:27.08,Default,,0000,0000,0000,,but some of those offered opt out. Dialogue: 0,0:11:28.45,0:11:31.70,Default,,0000,0000,0000,,This sort of intentional\Nyet incomplete random assignment Dialogue: 0,0:11:31.70,0:11:33.37,Default,,0000,0000,0000,,is surprisingly common. Dialogue: 0,0:11:33.37,0:11:36.32,Default,,0000,0000,0000,,Even randomized clinical trials\Nhave this feature. Dialogue: 0,0:11:37.13,0:11:40.05,Default,,0000,0000,0000,,IV solves the problem\Nof non-random take-up Dialogue: 0,0:11:40.05,0:11:42.53,Default,,0000,0000,0000,,in lotteries or clinical research. Dialogue: 0,0:11:43.05,0:11:46.72,Default,,0000,0000,0000,,But lotteries are not the only source\Nof compelling instruments. Dialogue: 0,0:11:46.92,0:11:49.12,Default,,0000,0000,0000,,Many causal questions\Ncan be addressed Dialogue: 0,0:11:49.12,0:11:50.76,Default,,0000,0000,0000,,by naturally occurring Dialogue: 0,0:11:50.76,0:11:53.83,Default,,0000,0000,0000,,as good as randomly\Nassigned variation. Dialogue: 0,0:11:54.73,0:11:56.92,Default,,0000,0000,0000,,Here's a causal question for you: Dialogue: 0,0:11:56.92,0:11:59.45,Default,,0000,0000,0000,,Do women who have children\Nearly in their careers Dialogue: 0,0:11:59.45,0:12:01.65,Default,,0000,0000,0000,,suffer a substantial earnings penalty Dialogue: 0,0:12:01.65,0:12:02.65,Default,,0000,0000,0000,,as a result? Dialogue: 0,0:12:02.65,0:12:04.97,Default,,0000,0000,0000,,After all, women earn less than men. Dialogue: 0,0:12:05.57,0:12:08.51,Default,,0000,0000,0000,,We could, of course, simply compare\Nthe earnings of women Dialogue: 0,0:12:08.51,0:12:10.89,Default,,0000,0000,0000,,with more and fewer children. Dialogue: 0,0:12:10.89,0:12:14.19,Default,,0000,0000,0000,,But such comparisons are fraught\Nwith selection bias. Dialogue: 0,0:12:14.81,0:12:17.40,Default,,0000,0000,0000,,If only we could\Nrandomly assign babies Dialogue: 0,0:12:17.40,0:12:19.09,Default,,0000,0000,0000,,to different households. Dialogue: 0,0:12:19.09,0:12:22.13,Default,,0000,0000,0000,,Yeah, right,\Nsounds pretty fanciful. Dialogue: 0,0:12:22.47,0:12:26.71,Default,,0000,0000,0000,,Our next IV story -- fantastic\Nand not fanciful -- Dialogue: 0,0:12:26.71,0:12:30.23,Default,,0000,0000,0000,,illustrates an amazing,\Nnaturally occurring instrument Dialogue: 0,0:12:30.23,0:12:31.92,Default,,0000,0000,0000,,for family size. Dialogue: 0,0:12:33.32,0:12:34.55,Default,,0000,0000,0000,,♪ [] ♪ Dialogue: 0,0:12:34.55,0:12:38.20,Default,,0000,0000,0000,,- [Instructor] You're on your way\Nto mastering econometrics. Dialogue: 0,0:12:38.20,0:12:40.17,Default,,0000,0000,0000,,Make sure this video sticks Dialogue: 0,0:12:40.17,0:12:42.64,Default,,0000,0000,0000,,by taking a few\Nquick practice questions. Dialogue: 0,0:12:42.89,0:12:46.34,Default,,0000,0000,0000,,Or, if you're ready,\Nclick for the next video. Dialogue: 0,0:12:46.53,0:12:50.20,Default,,0000,0000,0000,,You can also check out\NMRU's website for more courses, Dialogue: 0,0:12:50.20,0:12:52.03,Default,,0000,0000,0000,,teacher resources, and more. Dialogue: 0,0:12:52.29,0:12:53.77,Default,,0000,0000,0000,,♪ [] ♪