Causal Inference for Data Science

class: center, middle, inverse, title-slide

# Causal Inference for Data Science
## ITAM Short Workshop
### Zhe Zhang, Mathew Kiang, Monica Alexander
### March 16, 2017

---

name: inverse
layout: true
class: center, middle, inverse

---
# Impact of Information on Markets

- "The Digital Provide: Information (Technology), Market Performance, and Welfare in the South Indian Fisheries Sector" (Robert Jensen, QJE 2007)

- Special:
  - entry of cell phones is arguably exogenous to the fisherman and the fish markets
  - usual studies of impact of cell phones is macroeconomic and difficult to tease out cause and effect
  - important to look at the impact of information enabled by technology

![jensen_phones](./assets/jensen_phone_adoption.png)
![jensen_prices](./assets/jensen_price_changes.png)

---
# Impact of Additional Education?

![iv_AA_labor_laws](./assets/ang_acem_labor_laws_iv.png)

---
# Amazon Q&A Impact

- Newegg, etc.
- Imagine you work at Amazon
- Product specific controls by comparing two platforms (one with Q&A, one without)
- Scraped the data, comparing the sales ranks, controls
- Have to argue that Platform 2 & Amazon have similar time trends (e.g. what about Amazon sales being all on the first day and dropping off after that, while Platform 2 is stable) and track each other similarly
- Using video games, they can identify "new" product entries and measure uniqueness across platforms.

---
# Michael Luca

- impact of quality disclosure and rankings?
- impact of Yelp reviews?

---
# Wikipedia Effect?

- What is effect of Wikipedia? Hard Question; Georgia Tech econ on translation variety

---
# AirBnB in Austin

---
# Impact of Yelp in Seattle

---
# Diff in Diff on Impact of Fake Hotel Reviews

- initial consistent difference between TripAdvisors and Expedia
- use 2nd variation on nearby value of independent hotels

---
# Reddit Random Voting

(Science paper, Muchnik et al 2013)
Effects, very diverse
Do we believe it though? It's not clear that down-voting can be triggered though.
Also upvotes affect the chance a posting is viewed, difficult.

---
# Marketplace RCTs

Testing treatment groups can affect availability for the other control group
Or affecting only a portion of a seller's categories of goods

---
# Sources

- Cameron & Triveti textbook
- Mostly Harmless Econometrics
- Osea Giuntella, slides