analysis.qmd

---
title: "social-rl prelim analysis"
format: html
---

Packages.

```{r}
library(here)
library(tidyverse)
library(janitor)
```

Load data. 

```{r}
run <- "exp1"
```

```{r}
if (run == "pilota"){
  d_raw <- read_csv(here("data","amg_gamelogs_pilota.csv")) |>
  clean_names() 
} else if (run == "pilotb") {
  d_raw <- read_csv(here("data","amg_gamelogs_pilotb.csv")) |>
  clean_names() 
} else if (run == "exp1") {
  d_raw <- read_csv(here("data","amg_gamelogs_experiment1.csv")) |>
  clean_names() 
}
d_raw

```
Clean up data. 

```{r}
if (run == "pilota"){
    d <- d_raw |>
    filter(!is.na(condition_index)) |>
    select(-id, -v)
} else if (run == "pilotb") { 
  d <- d_raw |>
  filter(prolific_id!="Julio_test") |>
  filter(uid!="lpz173o40.k42j9exeqx") |>
  filter(!is.na(condition_index)) |>
  select(-id, -v, -payoffs, -payoffs_id, -policy)
  
  # Remove users who refresh their web-browser
  ids_to_remove <- d |>
    filter(refresh_count > 0) |>
    distinct(prolific_id) |>
    pull(prolific_id)
  
  d <- d |>
    filter(!prolific_id %in% ids_to_remove)
  
  
} else if (run == "exp1") { 
  prolific_ids_to_remove <- c('650c562d8a6ee3fcc26ee7ba', unique(filter(d_raw, refresh_count > 0)$prolific_id) )
  uids_to_remove <- c("lq1fvt8g0.ohk0zzv23oa", "lq1guank0.7qsa7lr1yjq", "lq1gdwe80.5ogj3d9flwk")
  
  d <- d_raw |>
    filter(!prolific_id %in% prolific_ids_to_remove) |>
    filter(!uid %in% uids_to_remove) |>
    filter(!is.na(condition_index)) |>
    select(-id, -v, -payoffs, -payoffs_id, -policy)
}

d  
```

```{r}
# Assuming your dataframe is named df
# Check for condition_count_index
df <- d

uids_condition_count <- df |> 
  group_by(uid, condition_index, block_index) |> 
  summarise(unique_condition_count = n_distinct(condition_count_index)) |> 
  filter(unique_condition_count != 2) |> 
  pull(uid)

# Check for block_index
uids_block <- df |> 
  group_by(uid, condition_index) |> 
  summarise(unique_blocks = n_distinct(block_index)) |> 
  filter(unique_blocks != 3) |> 
  pull(uid)

# Check for condition_index
uids_condition <- df |> 
  group_by(uid) |> 
  summarise(unique_conditions = n_distinct(condition_index)) |> 
  filter(unique_conditions != 4) |> 
  pull(uid)

# Combine all uids and get unique values
uids_not_satisfying <- unique(c(uids_condition_count, uids_block, uids_condition))

# View the uids that do not satisfy the criteria
print(uids_not_satisfying)

d <- d |>
    filter(!uid %in% uids_not_satisfying)


```



# Descriptives

```{r}
d |>
  select(prolific_id) |>
  distinct() |>
  count()
```
```{r}
d |>
  group_by(uid) |>
  summarise(time = max(timestamp) - min(timestamp))
```

# Main analysis

Aggregation. 
```{r}
participant_block_scores <- d |> 
  group_by(uid, condition_index, teacher, block_index, game_type) |>
  summarise(total_reward = max(total_reward))
participant_block_scores
```

```{r}
mean_block_scores <- participant_block_scores |>
  group_by(uid, condition_index) |>
  arrange(uid, condition_index, block_index, game_type) |>
  mutate(teacher_condition = teacher[!is.na(teacher)][1], 
         block_condition = reduce(game_type, paste)) |>
  group_by(teacher_condition, block_condition, block_index, game_type) |>
  select(teacher_condition, block_condition, block_index, game_type, condition_index, total_reward)

mean_block_scores$method <- "participants"
  
mean_block_scores
write_csv(mean_block_scores, here("data","exp1_human_performance.csv"))
```


```{r}
mean_block_scores <- participant_block_scores |>
  group_by(uid, condition_index) |>
  arrange(uid, condition_index, block_index, game_type) |>
  mutate(teacher_condition = teacher[!is.na(teacher)][1], 
         block_condition = reduce(game_type, paste)) |>
  group_by(teacher_condition, block_condition, block_index, game_type) |>
  summarize(mean_reward = mean(total_reward),
            sd_reward = sd(total_reward), 
            sem_reward = sd(total_reward)/sqrt(length(total_reward)),
            ci_reward = 1.96 * sem_reward)
mean_block_scores
```


```{r}
ggplot(mean_block_scores, 
       aes(x = block_index, y = mean_reward, col = teacher_condition)) +
  facet_wrap(~block_condition) + 
  geom_pointrange(aes(shape = game_type, ymin = mean_reward - ci_reward, 
                      ymax = mean_reward + ci_reward), 
                  position = position_dodge(width = .1)) +
  labs( x = "block index", y = "mean reward") +
  geom_line()
```