这个作业是对不同国家的餐饮计划进行差异化分析、评估
Economics 691-06: Assignment 4
1.（5分）使用事件研究方法，计算治疗效果（以及
标准错误）。请注意您在确定预处理时间方面所做的任何选择
期。
2.（10分）现在，我们准备进行差异分析。为此，请检查
美国，加拿大和澳大利亚的平行趋势假设。面向哪些国家
这个假设是好是坏？
3.（5分）使用差异差异方法和您的评估
之前的问题，请计算治疗效果（以及标准误差）。
问题2：准实验方法，第二部分
Blue Apron是另一家与Hello Fresh竞争的餐食制备公司，并且仅在美国运营。假设在这个程式化的问题中，它过去也有一个单一的用餐计划；
最近制定了一项低钠计划，希望与消费者进行测试。不同于Hello Fresh，
Blue Apron过去没有进行过满意度调查，因此不能依赖历史数据。上
好的一面是，与“ Hello Fresh”不同，Blue Apron确实具有后勤基础设施
在同一国家/地区提供两种不同的用餐计划。
蓝围裙由于接受低钠食品的消费者的PR风险而不想随机分组
未经他们同意而用餐。这样，它将打开一个在线队列。 Blue Apron能够为排队的前10,000位客户提供低钠餐点；尽管它无法为
剩下的（他们继续接受常规饮食），因为蓝围裙后来意识到
有足够的低钠餐。 Blue Apron要求您使用此数据来估计治疗效果。
数据位于文件数据分配4 2.csv中。该文件有30,000行。每行对应一个
单独的用户，有四列。第一列是用户在队列中的位置，
其中NA表示用户选择根本不注册该队列。第二栏
使用者是否接受过治疗，即低钠餐（以1表示），还是继续
接受常规饮食（以0表示）。第三列是结果，即用户的
调查的满意度（从0到10）。第四列称为“弹出式窗口”，应放在
待到问题稍后再讲。
1. (5 points) Why can we not regress satisfaction on treatment for a valid treatment effect?
Explain this in concrete language relevant to this scenario.
2. You decide to pursue a regression discontinuity design, in which you analyze consumers around
the queue cutoff of 10,000.
(a) (5 points) Using the same concrete language relevant to this scenario, why can we
compare consumers’ outcomes just above and below the cutoff to get a valid treatment
effect? In your answer, note any assumptions and whether you believe them to satisfied.
(You do not need to formally test the assumptions.)
(b) (5 points) Using the regression discontinuity methodology and all data, estimate the
treatment effect (along with standard errors).
(c) (5 points) Use the same methodology but restrict to data within a narrow band around
the cutoff: those with queue places between 9,000 and 11,000. Why might the result
differ, and which result do you trust more?
3. An engineering team at Blue Apron tells you that they actually were running a separate
experiment over this timeframe. Specifically, they were running pop-ups for some users,
which informed them about new developments at Blue Apron; and some pop-ups were indeed
related to these low-sodium meals and the queue for them. This is the fourth column –
“popups” – which tells if you users were receiving these pop-ups (indicated by 1) or not
(indicated by 0). You decide to pursue an instrumental variables approach, in which pop-ups
are the instrument.
(a) (10 points) Using the same concrete language relevant to this scenario, why might
pop-ups be a valid instrument for this problem? In your answer, note any assumptions
and whether you believe them to satisfied.
(b) (10 points) Using the instrumental variables methodology, estimate the treatment effect.
(c) (5 points) Using the bootstrap, compute the standard errors of your treatment effect.