# 00_generate_datasets.R ------------------------------------------------------ # MedEd Conclave 2026 · R & Reproducible Research Workshop # # Generates the three synthetic teaching datasets used across both days. # Nothing here is real. The structure, ranges and relationships are plausible # for an Indian medical college cohort, but no student, faculty member or # institution is represented. # # Run once from the project root: # source("R/00_generate_datasets.R") # # Outputs (written to data/): # meded_students.csv 240 rows · one row per student · clean # osce_stations.csv 1200 rows · one row per station · clean, long # faculty_survey_raw.csv 132 rows · one row per response · deliberately messy # codebook.csv · variable dictionary # ----------------------------------------------------------------------------- library(dplyr) library(tidyr) library(readr) library(stringr) set.seed(20260909) # workshop date; change this and every number below changes if (!dir.exists("data")) dir.create("data") n_stu <- 240 # 1. Student cohort ----------------------------------------------------------- colleges <- c("Govt Medical College, Thrissur", "Govt Medical College, Kozhikode", "Govt Medical College, Alappuzha") students <- tibble( student_id = sprintf("S%03d", 1:n_stu), batch = sample(c("2022", "2023", "2024"), n_stu, replace = TRUE, prob = c(0.3, 0.35, 0.35)), college = sample(colleges, n_stu, replace = TRUE, prob = c(0.4, 0.35, 0.25)), sex = sample(c("Female", "Male"), n_stu, replace = TRUE, prob = c(0.54, 0.46)), age = round(rnorm(n_stu, mean = 21.4, sd = 1.3)), rural_background = sample(c("Rural", "Urban"), n_stu, replace = TRUE, prob = c(0.42, 0.58)), # Randomised teaching arm - the comparison that runs through the workshop teaching_arm = rep(c("Traditional lecture", "Flipped classroom"), length.out = n_stu) |> sample() ) |> mutate( # Baseline ability drives everything downstream admission_score = round(rnorm(n_stu, mean = 68, sd = 9), 1), attendance_pct = round(pmin(100, pmax(45, rnorm(n_stu, 82, 9))), 1), pretest = round(pmin(100, pmax(0, 0.45 * admission_score + 0.10 * attendance_pct + rnorm(n_stu, 22, 7) )), 1), # Flipped classroom adds ~5 marks on average; attendance modifies it arm_effect = if_else(teaching_arm == "Flipped classroom", 5.2, 0), posttest = round(pmin(100, pmax(0, pretest + arm_effect + 0.06 * attendance_pct + rnorm(n_stu, 6, 5) )), 1), gain = round(posttest - pretest, 1), # Study hours per week, right-skewed study_hours = round(pmin(40, rgamma(n_stu, shape = 4, scale = 2.6)), 1), # Burnout: Copenhagen-style 0-100, inversely related to attendance burnout_score = round(pmin(100, pmax(0, 70 - 0.35 * attendance_pct + rnorm(n_stu, 8, 11) )), 1), burnout_band = cut( burnout_score, breaks = c(-Inf, 35, 60, Inf), labels = c("Low", "Moderate", "High") ), satisfaction = sample(1:5, n_stu, replace = TRUE, prob = c(.05, .12, .28, .35, .20)) ) |> select(-arm_effect) # Realistic missingness: some students skipped the post-test or the burnout scale students <- students |> mutate( posttest = if_else(row_number() %in% sample(seq_len(n_stu), 11), NA_real_, posttest), gain = if_else(is.na(posttest), NA_real_, gain), burnout_score = if_else(row_number() %in% sample(seq_len(n_stu), 7), NA_real_, burnout_score), burnout_band = if_else(is.na(burnout_score), NA_character_, as.character(burnout_band)), passed = if_else(!is.na(posttest) & posttest >= 50, "Yes", "No") ) write_csv(students, "data/meded_students.csv") # 2. OSCE stations (long format, for joins and grouped summaries) ------------- stations <- tribble( ~station, ~station_name, ~domain, "ST1", "History taking", "Communication", "ST2", "Clinical examination", "Clinical skill", "ST3", "Procedure: IV access", "Procedural skill", "ST4", "Counselling", "Communication", "ST5", "Data interpretation", "Cognitive" ) examiners <- sprintf("EX%02d", 1:12) osce <- expand_grid( student_id = students$student_id, station = stations$station ) |> left_join(stations, by = "station") |> left_join(select(students, student_id, admission_score, teaching_arm), by = "student_id") |> mutate( # Station difficulty: procedures are marked harder than communication difficulty = case_when( domain == "Procedural skill" ~ -1.8, domain == "Cognitive" ~ -0.9, TRUE ~ 0.4 ), arm_bonus = if_else(teaching_arm == "Flipped classroom" & domain == "Communication", 0.9, 0), score = round(pmin(20, pmax(0, 8 + 0.09 * (admission_score - 68) + difficulty + arm_bonus + rnorm(n(), 0, 2.1) ))), max_score = 20, examiner = sample(examiners, n(), replace = TRUE), attempt = 1L ) |> select(student_id, station, station_name, domain, examiner, score, max_score, attempt) write_csv(osce, "data/osce_stations.csv") # 3. Faculty development survey (deliberately messy) -------------------------- # Used for the data-cleaning parts of the dplyr / EDA session. The mess is # intentional and mirrors what actually arrives from a Google Form export. n_f <- 120 likert <- c("Strongly disagree", "Disagree", "Neutral", "Agree", "Strongly agree") faculty <- tibble( `Timestamp` = sample( c("2026-03-14", "14/03/2026", "14-03-2026", "2026/03/15"), n_f, replace = TRUE), `Respondent ID` = sprintf("F%03d", 1:n_f), `Department ` = sample( c("Community Medicine", "community medicine", "Physiology", "Paediatrics", "Pediatrics", " Anatomy", "Medical Education Unit"), n_f, replace = TRUE), `Years of Experience` = as.character(pmax(0, round(rgamma(n_f, 3, 0.25)))), `Ever used R?` = sample(c("Yes", "yes", "No", "no", "N", ""), n_f, replace = TRUE, prob = c(.12, .06, .5, .2, .1, .02)), `Confidence with stats (1-10)` = as.character(sample(1:10, n_f, replace = TRUE)), `Q1: I can reproduce my own analysis six months later` = sample(likert, n_f, replace = TRUE, prob = c(.18, .27, .25, .22, .08)), `Q2: I keep raw and cleaned data separate` = sample(likert, n_f, replace = TRUE, prob = c(.12, .22, .26, .28, .12)), `Q3: I would use AI tools to draft analysis code` = sample(likert, n_f, replace = TRUE, prob = c(.09, .15, .24, .34, .18)), `Hours spent on data cleaning per paper` = sample(c("2", "5", "10", "20", "40", "unknown", "NA", ""), n_f, replace = TRUE), `Comments` = sample( c("", "Very useful", " needs more hands on time ", "NA", "Please share slides", "-", "Would like a follow up session"), n_f, replace = TRUE) ) # Add 12 duplicated rows and scatter them - a real export always has some faculty_messy <- bind_rows(faculty, slice_sample(faculty, n = 12)) |> slice_sample(prop = 1) write_csv(faculty_messy, "data/faculty_survey_raw.csv", na = "") # 4. Codebook ----------------------------------------------------------------- codebook <- tribble( ~dataset, ~variable, ~type, ~description, "meded_students.csv", "student_id", "character", "Unique student identifier", "meded_students.csv", "batch", "character", "Year of admission", "meded_students.csv", "college", "character", "Medical college (3 sites)", "meded_students.csv", "sex", "character", "Female / Male as self-reported", "meded_students.csv", "age", "numeric", "Age in completed years", "meded_students.csv", "rural_background", "character", "Rural / Urban domicile", "meded_students.csv", "teaching_arm", "character", "Traditional lecture / Flipped classroom", "meded_students.csv", "admission_score", "numeric", "Entrance percentile at admission (0-100)", "meded_students.csv", "attendance_pct", "numeric", "Attendance across the module (%)", "meded_students.csv", "pretest", "numeric", "Knowledge test before the module (0-100)", "meded_students.csv", "posttest", "numeric", "Knowledge test after the module (0-100); 11 missing", "meded_students.csv", "gain", "numeric", "posttest - pretest", "meded_students.csv", "study_hours", "numeric", "Self-reported study hours per week", "meded_students.csv", "burnout_score", "numeric", "Burnout scale 0-100, higher = worse; 7 missing", "meded_students.csv", "burnout_band", "character", "Low (<35) / Moderate (35-60) / High (>60)", "meded_students.csv", "satisfaction", "numeric", "Course satisfaction, 1-5 Likert", "meded_students.csv", "passed", "character", "Yes if posttest >= 50", "osce_stations.csv", "student_id", "character", "Links to meded_students.csv", "osce_stations.csv", "station", "character", "Station code ST1-ST5", "osce_stations.csv", "station_name", "character", "Station description", "osce_stations.csv", "domain", "character", "Communication / Clinical / Procedural / Cognitive", "osce_stations.csv", "examiner", "character", "Examiner code EX01-EX12", "osce_stations.csv", "score", "numeric", "Station score, 0-20", "osce_stations.csv", "max_score", "numeric", "Maximum possible (20)", "faculty_survey_raw.csv", "(various)", "character", "Raw Google Form export: untidy names, mixed date formats, duplicates, inconsistent categories" ) write_csv(codebook, "data/codebook.csv") # 5. Report ------------------------------------------------------------------- cat("\nDatasets written to data/\n") cat(sprintf(" meded_students.csv %4d rows x %2d cols\n", nrow(students), ncol(students))) cat(sprintf(" osce_stations.csv %4d rows x %2d cols\n", nrow(osce), ncol(osce))) cat(sprintf(" faculty_survey_raw.csv %4d rows x %2d cols\n", nrow(faculty_messy), ncol(faculty_messy))) cat(sprintf(" codebook.csv %4d rows\n\n", nrow(codebook))) cat("Seed: 20260909. Re-running reproduces these files byte for byte.\n")