01-data-prep.Rmd

---
title: 'STUMBL: Prepare data'
author: "George Kinnear"
date: "26/03/2021"
always_allow_html: true
output:
  github_document:
    html_preview: false
  html_document:
    toc: false
---


```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = FALSE, message = FALSE, warning = FALSE)
library(tidyverse)
library(knitr)
library(kableExtra)
library(janitor)

# Set preferred styling
theme_set(theme_minimal())
colours_fac_vs_not = c("No FAC" = "#28c4d8", "FAC" = "#af003d")

basic_kable = function(df) {
  df %>% 
    kable() %>%
    kable_styling(bootstrap_options = "striped", full_width = F)
}
```

# Data sources

## Diagnostic Test response data

```{r message=FALSE, warning=FALSE}

diagtest_files <- fs::dir_ls(path = "data-ANON", regexp = ".*diagtest.*")

diagtest_results_all <- tibble(path = diagtest_files) %>% 
  # add columns describing the course and cohort from parsing the filename
  mutate(
    cohort = str_extract(path, "(?<=ANON_)(.*)(?=diagtest)"),
    diet = case_when(
      str_detect(path, "Sep") ~ "Pre",
      str_detect(path, "Jan") ~ "Post"
    )
  ) %>% 
  mutate(diet = fct_relevel(diet, "Pre", "Post")) %>% 
  # read in the file contents
  mutate(
    csv_contents = map(path, read_csv, na = c("-"))
  ) %>% 
  # unpack all the results and tidy up
  unnest(cols = c(csv_contents)) %>% 
  janitor::clean_names() %>% 
  # csv path is no longer needed
  select(-path) %>% 
  rename_with(~ gsub("_5_00", "", .x, fixed = TRUE)) %>% 
  rename(total = grade_100_00)

```
Number of responses:

```{r}
diagtest_results_all %>% 
  group_by(cohort, diet) %>% 
  summarise(
    n = n(),
    diagtest_mean = mean(total, na.rm = TRUE),
    diagtest_sd = sd(total, na.rm = TRUE),
    .groups = "drop"
  ) %>% 
  basic_kable() %>% 
  collapse_rows(columns = 1:2, valign = "top")
```

## Demographic data

We have demographic data available -- while we don't make use of this in analyses, we need to use this to determine which students are enrolled on which courses in each year, particularly for 2020/21 CAP where there are no results available for that yet.

```{r}
demographics <- read_csv("data-ANON/ANON_demographics.csv") %>% 
  # There are very few class only or exam only enrolments, and they are not relevant for us
  filter(!str_detect(course_mode_of_study, "Class Only|Exam only")) %>% 
  # neaten up the values
  mutate(
    cohort = case_when(
      course_year_code == "2018/9" ~ "1819",
      course_year_code == "2019/0" ~ "1920",
      course_year_code == "2020/1" ~ "2021",
    ),
    course = case_when(
      str_detect(course_code_title, "MATH08057") ~ "ILA",
      str_detect(course_code_title, "MATH08058") ~ "CAP",
      str_detect(course_code_title, "MATH07003") ~ "FAC",
    )
  )

demographics %>%
  select(AnonID, cohort, course) %>% 
  distinct() %>% 
  tabyl(course, cohort) %>%
  basic_kable()
```

## All course results

```{r}
results_all <-
  read_csv("data-ANON/ANON_all_course_results.csv") %>%
  janitor::clean_names() %>%
  select(anon_id, course_year, course_code, course_name, assessment_mark) %>% 
  filter(course_code %in% c("MATH07003", "MATH08057", "MATH08058")) %>% 
  mutate(
    cohort = case_when(
      course_year == "2018/9" ~ "1819",
      course_year == "2019/0" ~ "1920",
      course_year == "2020/1" ~ "2021",
    ),
    course = case_when(
      course_code == "MATH07003" ~ "FAC",
      course_code == "MATH08057" ~ "ILA",
      course_code == "MATH08058" ~ "CAP",
    )
  ) %>%
  select(anon_id, cohort, course, assessment_mark) %>% 
  # filter out the small number of students on CAP 2021 who score 0 already because they withdrew mid-semester
  filter(!(course == "CAP" & cohort == "2021"))

results_all %>% 
  tabyl(cohort, course) %>%
  adorn_totals(c("row", "col")) %>%
  basic_kable()
```

Combining these with Diagnostic Test pre/post results:

```{r}
diagtest_wide <- diagtest_results_all %>% 
  select(cohort, diet, anon_id, total) %>% 
  semi_join(demographics, by = c("anon_id" = "AnonID")) %>% 
  pivot_wider(
    names_from = diet,
    values_from = total,
    values_fn = max
  ) %>% 
  mutate(
    diagtest_group = case_when(
      !is.na(Pre) & !is.na(Post) ~ "Both",
      !is.na(Pre) ~ "Pre only",
      !is.na(Post) ~ "Post only",
      TRUE ~ "Neither"
    )
  )

results_all_wide <- results_all %>% 
  pivot_wider(
    names_from = course,
    values_from = assessment_mark,
    values_fn = max
  )

diagtest_plus_results_wide <- diagtest_wide %>% 
  full_join(results_all_wide, by = c("cohort", "anon_id")) %>% 
  mutate(took_FAC = as.factor(ifelse(is.na(FAC), "No FAC", "FAC")))

diagtest_plus_results_summary <- diagtest_plus_results_wide %>% 
  group_by(cohort) %>% 
  rename(student = anon_id) %>% 
  summarise(
    across(
      c(student, Post, FAC, ILA, CAP),
      list(
        total = ~ sum(!is.na(.)),
        withPre = ~ sum(!is.na(.) & !is.na(Pre)),
        pc = ~ sum(!is.na(.) & !is.na(Pre))/sum(!is.na(.))
      ),
      .names = "{.col}_{.fn}"
    )
  )

n_summary_reshaped <- diagtest_plus_results_summary %>% 
  pivot_longer(
    cols = contains("_"),
    names_to = "field",
    values_to = "value"
  ) %>% 
  separate(field, into = c("item", "detail"), sep = "_") %>% 
  filter(!(item =="CAP" & cohort == "2021")) %>% 
  mutate(item = str_replace(item, "student", "All students")) %>% 
  mutate(item = str_replace(item, "Post", "Took Post-test")) %>% 
  pivot_wider(
    names_from = "detail",
    values_from = "value"
  ) %>% 
  mutate(
    withPre = paste0(withPre, " (", round(pc*100,0), "%)")
  ) %>% 
  select(-pc) %>% 
  pivot_wider(
    id_cols = c(cohort, item),
    names_from = "cohort",
    values_from = c(total, withPre),
    names_glue = "{cohort}_{.value}"
  ) %>% 
  select(item, sort(current_vars()))

options(knitr.kable.NA = '-')
n_summary_reshaped %>% 
  kable(
    col.names = c("", rep(c("N", "Took Pre-test"), 3)),
    booktabs = TRUE,
    caption = "Table 3 in the paper",
    #format = "latex"
  ) %>%
  kable_styling(bootstrap_options = "striped", full_width = F) %>%
  add_header_above(c(" " = 1, "2018/19" = 2, "2019/20" = 2, "2020/21" = 2))


```

## Results distributions

Histograms of results in each of the three courses:

```{r}
results_all %>% 
  bind_rows(
    diagtest_results_all %>% select(cohort, course = diet, anon_id, assessment_mark = total) %>% 
      mutate(course = paste0(course, "-test"))
  ) %>% 
  mutate(course = fct_relevel(course, "Pre-test", "Post-test", "FAC", "ILA", "CAP")) %>% 
  mutate(cohort = paste0("20", str_sub(cohort, 1, 2), "/", str_sub(cohort, 3, 4))) %>% 
  ggplot(aes(x = assessment_mark)) +
  #geom_density() +
  geom_histogram(binwidth = 5, colour = "white", size = 0.1) +
  facet_grid(cols = vars(cohort), rows = vars(course), scales = "free", switch = "y") +
  labs(x = "Course result", y = "") +
  theme(
    strip.text.y.left = element_text(angle = 0),
    strip.placement = "outside"
  )
ggsave("FIG_course_results_histogram.pdf", width = 20, height = 18, units = "cm")

```



# Students with pre-post data

For pre-post analyses, we restrict to students who were enrolled on one of FAC/ILA/CAP (i.e. to exclude the many engineering students who only take the diagnostic test in September).


```{r}
diagtest_wide <- diagtest_results_all %>% 
  select(cohort, diet, anon_id, total) %>% 
  semi_join(demographics, by = c("anon_id" = "AnonID")) %>% 
  pivot_wider(
    names_from = diet,
    values_from = total,
    values_fn = max
  ) %>% 
  mutate(
    data_group = case_when(
      !is.na(Pre) & !is.na(Post) ~ "Both",
      !is.na(Pre) ~ "Pre only",
      !is.na(Post) ~ "Post only",
      TRUE ~ "Neither"
    )
  )

```

We assemble the available data for exploring the ""FAC Effect"", saving it as `data-ANON/ANON_student-data.csv`.

```{r assemble-fac-effect-data}

measures = tibble("meas" = c("Post", "ILA", "CAP")) %>% 
  mutate(
    meas_data_all = map(meas, function(meas) diagtest_plus_results_wide %>% 
      select(cohort, anon_id, took_FAC, Pre, outcome = meas) %>%
      filter(!is.na(outcome))
      #select(cohort, anon_id, took_FAC, Pre, meas) %>%
      #filter(!is.na(!!as.symbol(meas)))
      ),
    N_total = map_int(meas_data_all, nrow),
    meas_data = map(meas_data_all, function(dat) dat %>% 
        filter(!is.na(Pre))
      ),
    N_with_Pre = map_int(meas_data, nrow),
    pc = N_with_Pre / N_total
  )

measures %>% 
  select(meas, contains("N"), pc)  %>%
  basic_kable()

diagtest_plus_results_wide %>% 
  select(-diagtest_group) %>% 
  write_csv("data-ANON/ANON_student-data.csv")
  
```