这是一篇来自美国的关于因果性推断的作业代写限时测试

Instructions

You should submit your writeup (as a knitted .pdf along with the accompanying .rmd file) to the course website before 11:59pm EST on Saturday, August 20th. Please upload your solutions as a .pdf file saved as Yourlastname Yourfirstname final.pdf. In addition, an electronic copy of your .rmd file (saved as Yourlastname Yourfirstname final.rmd) should accompany this submission.

Late finals will not be accepted, so start early and plan to finish early. Remember that exams often take longer to finish than you might expect.

This exam has 5 questions and is worth a total of 100 points. Show your work in order to receive partial credit. Also, we will not accept un-compiled .rmd files.

In general, you will receive points (partial credit is possible) when you demonstrate knowledge about the questions we have asked, you will not receive points when you demonstrate knowledge about questions we have not asked, and you will lose points when you make inaccurate statements (whether or not they relate to the question asked). Be careful, however, that you provide an answer to all parts of each question.

You may use your notes, books, and internet resources to answer the questions below. However, you are to work on the exam by yourself. You are prohibited from corresponding with any human being  regarding the exam (unless following the procedures below).

We will answer clarifying questions during the exam. We will not answer statistical or computational questions until after the exam is over. If you have a question, post it on Campuswire as  a private post, visible only to the instructional staff. If your question is a clarifying one, we will reply.

Problem 1 (20 points)

This problem will have you replicate and analyze the results from Moser and Voena’s 2012 AER paper on the impact of the World War I Trading with the Enemy Act on U.S. domestic invention.

The full citation is below Moser, P., & Voena, A. (2012). Compulsory licensing: Evidence from the trading with the enemy act. American Economic Review, 102(1), 396-427.

The premise of the study is to evaluate the effect that ”compulsory licensing” policy – that is, policies that permit domestic firms to violate foreign patents and produce foreign inventions without needing to obtain a license from the owner of the foreign patent – have on domestic invention. Does access to foreign inventions make domestic firms more innovative? The authors leverage an exogenous event in U.S. licensing policy that arose from World War I – the 1917 ”Trading with the Enemy Act” (TWEA) which permitted U.S. firms to violate patents owned by enemy-country firms. This had the consequence of effectively licensing all patents from German-owned firms to U.S. firms after 1918 (that is, from 1919 onward), allowing them to produce these inventions without paying for a license from the German-owned company.

The authors look specifically at domestic innovation and patent activity in the organic chemicals sector. They note that only some of the sub-classes of organic chemicals (as defined by the US Patent Office) received any compulsory licenses under the Trading with the Enemy Act while others did not.

They leverage this variation in exposure to the ”treatment” of compulsory licensing to implement a differences-in-differences design looking at domestic firm patent activity in each of these sub-classes(comparing sub-classes that were exposed to compulsory licensing to those that were unexposed).

The unit of the dataset is the sub-class/year (471,120 observations) of 7248 US Patent and Trade mark Office (USPTO) patent sub-classes over 65 years.

The dataset is patents.csv and the relevant variables are:

• uspto class – USPTO Patent Sub-Class (unit)
• grntyr – Year of observation (year)
• count usa – Count of patents granted to US-owned firms in the year
• count for – Count of patents granted to foreign-owned (non-US) firms in the year
• treat – Treatment indicator – Whether the patent sub-class received any German patents under TWEA (after 1918 when the policy went into effect) (Note that this is not an indicator for the overall treatment group (whether the unit ever received treatment) – it is only 1 after 1918 for units that receive treatment but is still 0 for those ”treated” units prior to the initiation of treatment)

Question A ( 5 points)

If you try to use a two-way fixed effects estimator on the dataset as it is, it will likely freeze up your computer as this is a very large dataset. We’ll instead first aggregate the data in a way that will let you use a simple difference-in-differences estimator to estimate the treatment effect.

Generate a point estimate for the average treatment effect of receiving treatment on the average annual count of US patents using a difference-in-differences estimator (using all post-treatment (1919-1939) and pretreatment (1875-1918) time periods. You should aggregate your data such that the outcome is the post-/pre-difference in the outcome (preferably using tidyverse functions like group by and summarize) and each row is a USPTO patent sub-class (rather than a sub-class/year observation) and use a difference-in-means estimator with the differenced outcome. Again, if you use 1m robust or even 1m with two-way fixed effects, your computer will likely freeze up as there are many FE parameters to estimate.

Provide a 95% robust confidence interval and interpret your point estimate. Do we reject the null of no treatment effect at the α = .05 level?

1918 (that is, from 1919 onward), allowing them to produce these inventions without paying for a license from the German-owned company.

The authors look specifically at domestic innovation and patent activity in the organic chemicals sector. They note that only some of the sub-classes of organic chemicals (as defined by the US Patent Office) received any compulsory licenses under the Trading with the Enemy Act while others did not.

They leverage this variation in exposure to the ”treatment” of compulsory licensing to implement a differences-in-differences design looking at domestic firm patent activity in each of these sub-classes (comparing sub-classes that were exposed to compulsory licensing to those that were unexposed).

The unit of the dataset is the sub-class/year (471,120 observations) of 7248 US Patent and Trade mark Office (USPTO) patent sub-classes over 65 years.

The dataset is patents.csv and the relevant variables are:

• uspto class – USPTO Patent Sub-Class (unit)
• grntyr – Year of observation (year)
• count usa – Count of patents granted to US-owned firms in the year
• count for – Count of patents granted to foreign-owned (non-US) firms in the year
• treat – Treatment indicator – Whether the patent sub-class received any German patents under TWEA (after 1918 when the policy went into effect) (Note that this is not an indicator for the overall treatment group (whether the unit ever received treatment) – it is only 1 after 1918 for units that receive treatment but is still 0 for those ”treated” units prior to the initiation of treatment)

Question A ( 5 points)

If you try to use a two-way fixed effects estimator on the dataset as it is, it will likely freeze up your computer as this is a very large dataset. We’ll instead first aggregate the data in a way that will let you use a simple difference-in-differences estimator to estimate the treatment effect.

Generate a point estimate for the average treatment effect of receiving treatment on the average annual count of US patents using a difference-in-differences estimator (using all post-treatment (1919-1939) and pretreatment (1875-1918) time periods. You should aggregate your data such that the outcome is the post-/pre-difference in the outcome (preferably using tidyverse functions like group by and summarize) and each row is a USPTO patent sub-class (rather than a sub-class/year observation) and use a difference-in-means estimator with the differenced outcome. Again, if you use 1m robust or even 1m with two-way fixed effects, your computer will likely freeze up as there are many FE parameters to estimate.

Provide a 95% robust confidence interval and interpret your point estimate. Do we reject the null of no treatment effect at the α = .05 level?