analysis.Rmd

---
title: "analysis"
output: html_document
---

```{r}
install.packages("dplyr")
install.packages("ggpubr")
install.packages("ggplot2")
install.packages("xtable")
install.packages("cowplot")

library(dplyr)
library(ggpubr)
library(ggplot2)
library(xtable)
library(cowplot)
```

Read the data to a dataframe

```{r}
data <- read.csv(
  "./data/activity_preprocessed.csv",
  header = TRUE,
  stringsAsFactors = FALSE
)
```


Convert abbreviated badge and activity identifiers to full-name-factors
```{r}
badge_names = c("Mortarboard", "Epic", "Legendary")
activity_names = c("Question", "Answer", "Comment")

data$badge_name = factor(
  data$badge_name,
  levels=c("m", "e", "l"),
  labels=badge_names
)

data$activity_name = factor(
  data$activity_name,
  levels=c("q", "a", "c"),
  labels=activity_names
)

```

Filter data to 4 weeks before and after getting the badge, sum activity counts
Note: week 0 starts at the time of badge receival

```{r}
data_before <- filter(data, data$week_offset >= -4 & data$week_offset < 0)
data_after <- filter(data, data$week_offset < 4 & data$week_offset >= 0)

data_before_agg <- data_before %>%
  group_by(user_id, badge_name, activity_name) %>%
  summarise(amount = sum(activity_count))

data_after_agg <- data_after %>%
  group_by(user_id, badge_name, activity_name) %>%
  summarise(amount = sum(activity_count))
```
Create lineplots

```{r}
for (badge in badge_names)
{
  lineplot_data <- data %>%
    filter(badge_name==badge) %>%
    group_by(week_offset, activity_name) %>%
    summarise(Avg = mean(activity_count_standardized, na.rm=TRUE))

  plot <- ggplot(data=lineplot_data,
         aes(x=week_offset, y=Avg, colour=activity_name)) +
         geom_line()+
         ggtitle(paste("Badge '", badge, "'", sep=""))+
         theme(plot.title = element_text(hjust = 0.5)) +
         xlab("Weeks after badge receival") +
         scale_x_continuous(breaks = scales::pretty_breaks(n = 10)) +
         ylab("Standardized activity count") +
         scale_y_continuous(breaks = scales::pretty_breaks(n = 10))
  plot <- plot + labs(colour = "Activity type")

  ggsave(
    paste("./results/", badge, "_line.pdf", sep=""),
    device="pdf",
    width=6,
    height=2.4,
  )
}
```

Check for normality

```{r}

par(mfrow=c(3,3))
for (badge in badge_names)
{
  for (activity in activity_names)
  {

    # filter full dataframes to required badge and activity
    before = filter(
      data_before_agg,
      badge_name == badge & activity_name == activity
    )$amount
    after = filter(
      data_after_agg,
      badge_name == badge & activity_name == activity
    )$amount

    # filter users with incomplete data
    invalid_mask = is.na(before) | is.na(after)
    after = after[!invalid_mask]
    before = before[!invalid_mask]

    qqnorm(after-before, main = c(badge, activity))
    qqline(after-before)
  }
}

```

Check prerequisite for Wilcoxon test (symmetric distribution)

```{r}
par(mfrow=c(3,3))
for (badge in badge_names)
{
  for (activity in activity_names)
  {
    # filter full dataframes to required badge and activity
    before = filter(
      data_before_agg,
      badge_name == badge & activity_name == activity
    )$amount
    after = filter(
      data_after_agg,
      badge_name == badge & activity_name == activity
    )$amount

    # filter users with incomplete data
    invalid_mask = is.na(before) | is.na(after)
    before = before[!invalid_mask]
    after = after[!invalid_mask]

    hist(after-before, breaks=50, main = c(badge, activity))

  }
}
```

Calculate statistics and perform paired Wilcoxon tests per activity and badge category

```{r}
# Initialize result vectors
badges <- c(
)
activities <- c()
medians_before <- c()
medians_after <- c()
wil_values <- c()
p_values_wilcox <- c()
sample_size <- c()
median_differences <- c()


for (badge in badge_names)
{
  for (activity in activity_names)
  {
    badges <- c(badges, badge)
    activities <- c(activities, activity)

    # filter full dataframes to required badge and activity
    before = filter(
      data_before_agg,
      badge_name == badge & activity_name == activity
    )$amount
    after = filter(
      data_after_agg,
      badge_name == badge & activity_name == activity
    )$amount

    # filter users with incomplete data
    invalid_mask = is.na(before) | is.na(after)
    before = before[!invalid_mask]
    after = after[!invalid_mask]

    # calculate medians and sample size
    medians_before <- c(medians_before, median(before))
    medians_after <- c(medians_after, median(after))
    sample_size <- c(sample_size, length(before))


    wtest <- wilcox.test(after, before, paired=TRUE, conf.int=TRUE)
    p_values_wilcox <- c(p_values_wilcox, wtest$p.value * 9)  # bonferroni correction
    wil_values <- c(wil_values, wtest$statistic)
    median_differences <- c(median_differences, wtest$estimate)

  }
}

results = data.frame(
  badge=badges,
  activity=activities,
  median.before=medians_before,
  median.after=medians_after,
  median.differences=median_differences,
  sample.size=sample_size,
  V=wil_values,
  p=p_values_wilcox
)


# store results in a csv for use in the report
write.csv(results, "./results/analysis_results_wilcox.csv", row.names=FALSE)
```

Create boxplots

```{r}
data_agg_combined = merge(data_before_agg, data_after_agg, by=c("user_id", "badge_name", "activity_name"))
data_agg_combined$diff = data_agg_combined$amount.y - data_agg_combined$amount.x
data_agg_combined
```

```{r}
boxplot_questions <- ggplot(
  data_agg_combined %>% filter(activity_name=="Question"),
  aes(x=badge_name, y=diff)
) +
  geom_boxplot() +
  facet_grid(. ~ activity_name) +
  coord_cartesian(ylim=c(-10, 10)) +
  geom_hline(yintercept = 0, color = "black")

boxplot_answers <- ggplot(
  data_agg_combined %>% filter(activity_name=="Answer"),
  aes(x=badge_name, y=diff)
) +
  geom_boxplot() +
  facet_grid(. ~ activity_name) +
  coord_cartesian(ylim=c(-60, 60)) +
  geom_hline(yintercept = 0, color = "black")

boxplot_comments <- ggplot(
  data_agg_combined %>% filter(activity_name=="Comment"),
  aes(x=badge_name, y=diff)
) +
  geom_boxplot() +
  facet_grid(. ~ activity_name) +
  coord_cartesian(ylim=c(-100, 100)) +
  geom_hline(yintercept = 0, color = "black")

cowplot::plot_grid(boxplot_questions, boxplot_answers, boxplot_comments, ncol=3)

ggsave(
  paste("./results/boxplots.pdf", sep=""),
  device="pdf",
  width=7,
  height=3.5,
)
```