---
title: "Causal Inference and Regression"
output: pdf_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
library(tidyverse)
library(knitr)
set.seed(04132021)
```

Causal inference can be characterized as a predictive problem, where the question is 

\vfill

Simulate and visualize data with two potential outcomes

\vfill

```{r, echo = F}
n <- 40
y0 <- rnorm(n)
y1 <- y0 + rnorm(n, mean = 2)
```

\vfill
```{r, echo = F}
dat <- tibble(id = factor(rep(1:n, 2)), response = c(y0, y1), 
              `potential outcome` = rep(c('y0','y1'), each = n))

dat %>% ggplot(aes(y = response, x = `potential outcome`)) + 
  geom_violin(draw_quantiles = c(.025, .5, .975)) + 
  geom_jitter(aes(color = id),  shape = rep(c(0:19),4)) +
  theme_bw() + theme(legend.position = 'none') +
  ggtitle('Potential Outcomes')
```

\newpage

Then randomly assign treatments to each unit and visualize differences
\vfill

```{r, echo = F}
treatment <- sample(rep(0:1, each = n/2))
sample_dat <- tibble(id = factor(1:n), response = y0, 
                     `potential outcome` = rep('y0', n)) %>% filter(treatment == 0) %>% bind_rows(
                    tibble(id = factor(1:n), response = y1, 
                     `potential outcome` = rep('y1', n)) %>% filter(treatment == 1)   
                     )
```

\vfill

```{r, echo = F}
sample_dat %>% ggplot(aes(y = response, x = `potential outcome`)) + 
  geom_violin(draw_quantiles = c(.025, .5, .975)) + 
  geom_jitter(aes(color = id)) + theme_bw() + theme(legend.position = 'none') +
    ggtitle('Factual Outcomes')

```


\newpage

#### Pre-treatment covariates

Consider a setting with: 


\vfill


\vfill



\vfill

\vfill

\vfill