Class 9: Diff-in-diff II

# In-person session 9

**March 7, 2024**

]

---

# Plan for today

.box-2.medium.sp-after-half[Quick questions]

.box-5.medium.sp-after-half[LLMs]

.box-1.medium.sp-after-half[Matching and IPW]

.box-6.medium.sp-after-half[Two-way fixed effects]

---

layout: false
name: models-designs
class: center middle section-title section-title-2 animated fadeIn

# Quick questions

---

---

.box-2.large[Can you combine design-based and model-based identification?]

.box-inv-2.medium[Like diff-in-diff with a DAG?]

---

.box-2.large[DIDID(IDIDID)?]

---

.box-2.medium[The effect of mandatory maternity benefits on wages]

.box-inv-2.medium[New Jersey implements policy; Pennsylvania doesn't]

.box-inv-2.medium[Only applies to married women who have kids]

---

.box-inv-2.medium[Married women 20–40 - single men/unmarried women/older women in NJ and PA]

---

.center[
<figure>
 <img src="img/09-class/ddd-cunningham-paper.webp" alt="Diff-in-diff-in-diff" title="Diff-in-diff-in-diff" width="50%">
</figure>
]

???

- <https://causalinf.substack.com/p/triple-differences-part-1>
- <https://causalinf.substack.com/p/triple-difference-part-3-triple-differences>

---

.box-2.large[The DAG from the test]

---

layout: false
name: llms
class: center middle section-title section-title-5 animated fadeIn

# LLMs

---

---

.box-5.medium[How have you used LLMs like ChatGPT?]

.box-5.medium[What worries have you had?]

---

.box-5.medium[Can we use LLMs like ChatGPT?]

.box-5.medium[How do we use them?]

.box-5.medium[Is it okay to use them?]

---

.box-5.medium[LLMs are not magical. They're stats.]

---

.center[
<iframe width="800" height="450" src="https://www.youtube.com/embed/67_aMPDk2zw" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
]

---

---

# So many ethical issues!

.pull-left.center[
.box-5.less-medium[Environmental concerns]

.tiny.center[["Environmental Impact of Large Language Models"](https://www.cutter.com/article/environmental-impact-large-language-models) and ["The mounting human and environmental costs of generative AI"](https://arstechnica.com/gadgets/2023/04/generative-ai-is-cool-but-lets-not-forget-its-human-and-environmental-costs/) and ["AI water footprint suggests that large language models are thirsty"](https://techhq.com/2023/04/ai-water-footprint-suggests-that-large-language-models-are-thirsty/)]

.box-5.less-medium[Racial and gender bias]

.tiny.center[["AI chatbots use racist stereotypes even after anti-racism training"](https://www.newscientist.com/article/2421067-ai-chatbots-use-racist-stereotypes-even-after-anti-racism-training/)]

.box-5.less-medium[Human toll]

.tiny.center[["OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic"](https://time.com/6247678/openai-chatgpt-kenya-workers/)]

]

.tiny.center[["ChatGPT Stole Your Work. So What Are You Going to Do?"](https://www.wired.com/story/chatgpt-generative-artificial-intelligence-regulation/) and ["Congress Wants Tech Companies to Pay Up for AI Training Data"](https://www.wired.com/story/congress-senate-tech-companies-pay-ai-training-data/) and ["ChatGPT can leak training data, violate privacy"](https://www.zdnet.com/article/chatgpt-can-leak-source-data-violate-privacy-says-googles-deepmind/)]

.box-5.less-medium[Ouroboros effect]

.tiny.center[["Meet the Serbian DJ Running an AI Clickbait Business"](https://www.wnycstudios.org/podcasts/otm/segments/serbian-dj-internet-ai-clickbait-on-the-media) and ["The Perfect Webpage"](https://www.theverge.com/c/23998379/google-search-seo-algorithm-webpage-optimization)]

.box-5.less-medium[Making stuff up]

.tiny.center[["AI models make stuff up. How can hallucinations be controlled?"](https://www.economist.com/science-and-technology/2024/02/28/ai-models-make-stuff-up-how-can-hallucinations-be-controlled)]
]

???

<https://www.thehairpin.com/>

---

.box-5.large[You need to figure out your own ethics.]

---

---

# LLMs and programming

.box-5.medium[GitHub Copilot specifically trained on code]

.box-5.medium[Works *surprisingly* well]

.box-inv-5.medium[But it's dangerous if you don't know what you're doing!]

---

# Copilot skills

.box-5.medium[Talking to Copilot requires special skills and practice!]

.box-5.medium[***Rep***roducible ***ex***amples!]

.small.center[[Reprex slides](https://datavizf23.classes.andrewheiss.com/slides/08-slides.html#15)]

.box-inv-5.smal[[GitHub Gists](https://gist.github.com/)]

---

# Things Copilot is good at

.box-5.medium[Explaining and annotating code]

.box-5.medium[Translating between languages]

.box-5.medium[Generating boilerplate/starter code]

.box-5.medium[Cleaning and rewriting code]

---

.box-5.huge[Demonstration!]

---

layout: false
name: ipw
class: center middle section-title section-title-1 animated fadeIn

# Matching and IPW

---

layout: false
name: twfe
class: center middle section-title section-title-6 animated fadeIn

# Two-way fixed effects (TWFE)

---

---

.box-6.medium[Two states: Alabama vs. Arkansas]

`$$\begin{aligned}
\text{Mortality}\ =&\ \beta_0 + \beta_1\ \text{Alabama} + \beta_2\ \text{After 1975}\ + \\
&\ \beta_3\ (\text{Alabama} \times \text{After 1975})
\end{aligned}$$`

---

.box-6.medium[All states: `Treatment == 1` if legal for 18-20-year-olds to drink]

`$$\text{Mortality}\ =\ \beta_0 + \beta_1\ \text{Treatment} + \beta_2\ \text{State} + \beta_3\ \text{Year}$$`

---

`$$\begin{aligned}
\text{Mortality}\ =&\ \beta_0 + \beta_1\ \text{Alabama} + \beta_2\ \text{After 1975}\ + \\
&\ \color{red}{\beta_3}\ (\text{Alabama} \times \text{After 1975})
\end{aligned}$$`

`$$\text{Mortality}\ =\ \beta_0 + \color{red}{\beta_1}\ \text{Treatment} + \beta_2\ \text{State} + \beta_3\ \text{Year}$$`

---

`$$\begin{aligned}
\text{Mortality}\ =&\ \beta_0 + \beta_1\ \text{Alabama} + \beta_2\ \text{After 1975}\ + \\
&\ \color{red}{\beta_3}\ (\text{Alabama} \times \text{After 1975})
\end{aligned}$$`

`$$\text{Mortality}\ =\ \beta_0 + \color{red}{\beta_1}\ \text{Treatment} + \beta_2\ \text{State} + \beta_3\ \text{Year}$$`

`$$\begin{aligned}
\text{Mortality}\ =\ & \beta_0 + \color{red}{\beta_1}\ \text{Treatment} + \beta_2\ \text{State} + \beta_3\ \text{Year}\ +\\
&\beta_4\ (\text{State} \times \text{Year})
\end{aligned}$$`

---

.center[
<figure>
 <img src="img/08-class/mm-tbl-5-2.png" alt="Mastering Metrics Table 5.2" title="Mastering Metrics Table 5.2" width="55%">
</figure>
]

---

`$$\begin{aligned}
\text{Donation rate}\ =&\ \beta_0 + \beta_1\ \text{California} + \beta_2\ \text{After Q22011}\ + \\
&\ \beta_3\ (\text{California} \times \text{After Q22011})
\end{aligned}$$`

$$
`\begin{aligned}
\text{Donation rate}\ =\ & \beta_0 + \color{red}{\beta_1}\ \text{Treatment}\ + \\
& \beta_2\ \text{State} + \beta_3\ \text{Quarter}
\end{aligned}`
$$

---

.box-6.large[What about this staggered treatment stuff?]

???

This is good for ethical reasons!

Blog post