By the end of today you can

• organise a project that runs from raw data to finished results;
• combine cleaning, description, modelling, and visualisation in one pipeline;
• trace the chain of reasoning from question to conclusion and limitation;
• report results as a thesis expects, following published standards;
• choose a method for a research question;
• recognise common challenges, and know where to learn more.

A thesis is an argument

The results chapter is the evidence.

Each claim rests on a chain: question, hypothesis, design, variables, analysis, result, conclusion, limitation.

It convinces when every link can be seen, and every number traced back to the raw data.

Rebuild it as one project

Analyses written over two years, as each method was learned, are scattered across scripts.

Before writing the results chapter: rebuild everything, from the raw export to the last table, as one project that runs from start to finish.

Then every number is right, and any examiner’s question can be answered by pointing to the code.

The shape of a research project

flowchart TB
  subgraph R1[" "]
    direction LR
    A["Question and design<br>Ch 5"] --> B["Import<br>Ch 1–2"] --> C["Clean<br>Ch 3"] --> D["Explore<br>Ch 4, 6"]
  end
  subgraph R2[" "]
    direction LR
    E["Test and model<br>Ch 7–10"] --> G["Report and share<br>Ch 17–18"]
    F["Predict and discover<br>Ch 11–16"] --> G
  end
  R1 --> R2

Not a straight line, but code makes going back easy.

Organising the project

wellbeing-thesis/
├── wellbeing-thesis.Rproj      the RStudio Project
├── README.txt                  what it is and how to run it
├── data-raw/wellbeing_raw.xlsx the export, never edited by hand
├── data/                       clean data, created by scripts
├── R/01-clean-data.R           raw export -> clean tables
├── R/02-analysis.R             models and figures
├── output/                     figures and tables, created by scripts
├── results.qmd                 the results chapter
├── references.bib
└── renv.lock                   package versions

Four principles

The downloadable Chapter 19 project has exactly this structure.

From raw export to clean data

responses <- raw |>
  filter(!str_detect(str_to_upper(`Q1_Student ID`), "^TEST")) |>
  select(-`Response ID`) |>
  distinct() |>
  rename(student_id = `Q1_Student ID`, workshop = `Workshop group`,
         considering_dropout = `Y1_Considered leaving?`) |>
  rename_with(~ item_names, .cols = Q11_1:Q11_22) |>
  mutate(across(all_of(item_names), ~ as.integer(na_if(.x, "99"))))

The cleaning of Chapter 3 in one script: it runs in seconds.

The semester records

semesters_clean <- responses |>
  select(student_id, matches("_S[1-4]$")) |>
  pivot_longer(-student_id, names_to = c(".value", "semester"),
               names_sep = "_S") |>
  mutate(sleep_hours = parse_number(str_replace(sleep_hours, ",", ".")),
         sleep_hours = if_else(sleep_hours > 24, NA, sleep_hours),
         gpa         = if_else(gpa > 4, NA, gpa)) |>
  filter(!if_all(gpa:wellbeing, is.na))

The most important line: the check

same(questionnaire_clean |> arrange(student_id),
     questionnaire |> arrange(student_id))
#> [1] TRUE
same(semesters_clean |> arrange(student_id, semester),
     semesters |> arrange(student_id, semester))
#> [1] TRUE

In a real project there is nothing to compare with: check counts, ranges, missing values, and a few rows (Chapters 3 and 6).

One analysis table

study <- students |>
  left_join(scores, join_by(student_id)) |>
  left_join(semesters |> filter(semester == 1), join_by(student_id))

One row per student: background, questionnaire scores, and first-semester records.

Helper functions instead of copy and paste

mean_sd <- function(x) sprintf("%.1f (%.1f)", mean(x, na.rm = TRUE),
                                               sd(x, na.rm = TRUE))
percent <- function(x) sprintf("%.0f%%", 100 * mean(x, na.rm = TRUE))

describe <- function(d) {
  tibble(Characteristic = c("Students", "Age, mean (SD)", ...),
         Value = c(nrow(d), mean_sd(d$age), ...))
}

The same describe() makes every column. Write a small function whenever you would copy and paste.

Table 1: the sample

Formatting p-values

format_p <- function(p) {
  if (p < 0.001) "p < .001"
  else paste("p =", sub("^0", "", sprintf("%.3f", p)))
}

Theses report p-values in a fixed style. A function applies it every time, without mistakes.

RQ3: the workshop and its persistence

workshop_model <- lmer(wellbeing ~ factor(semester) * workshop +
                         (time | student_id), data = panel)
gaps$wellbeing <- predict(workshop_model, newdata = gaps, re.form = NA)

Wellbeing was similar before the workshop (difference 0.4 points). Invited students’ wellbeing was 5.2 points higher in semester 2, narrowing to 4.1 in semester 3 and 2.6 in semester 4.

RQ5: explaining GPA

gpa_model <- lm(gpa ~ sleep_hours + study_hours + stress + support,
                data = study)
gpa_table <- tidy(gpa_model, conf.int = TRUE)

Each additional hour of sleep was associated with a GPA 0.11 points higher (95% CI 0.08 to 0.13, p < .001), holding study hours, stress, and support constant. The four predictors explained 25% of the variation (n = 587).

RQ9: considering dropout

dropout_model <- glm(
  I(considering_dropout == "Yes") ~ stress + support + financial_worry +
    employment + study_mode, data = study, family = binomial)
tidy(dropout_model, conf.int = TRUE, exponentiate = TRUE)

Each one-point increase in stress was associated with 3.6 times the odds of considering dropping out (95% CI 2.3 to 5.9); each point of support with 0.33 times the odds.

I() uses a condition as the outcome; exponentiate = TRUE gives odds ratios.

One figure for the thesis

patchwork: + side by side, / stacked, & applies a theme to all panels.

Saving the figure

(panel_a + panel_b) +
  plot_annotation(tag_levels = "A") &
  theme_minimal(base_size = 12)

ggsave(here::here("output", "figure-1.png"),
       width = 18, height = 8, units = "cm", dpi = 300)

At the size and resolution the thesis requires (Chapter 4).

A result is not yet a conclusion

Between them lie:

• the design that produced the data;
• how the variables were measured;
• the limits of both.

The chain of reasoning makes each link visible.

The chain of reasoning: question to analysis

The chain of reasoning: result to limitation

Only the workshop was randomised, so only its conclusion is causal.

The results chapter

The tables, figures, and sentences go into results.qmd, with every number written as inline code (Chapter 17).

When the data changes, or an examiner asks for a different model, the code changes and the chapter renders again.

Reporting standards

They list what readers need to judge a study, including effect sizes and confidence intervals. Check your draft against the relevant one.

Choosing a method: the goal

flowchart LR
  Q{What is the goal?} -->|Describe| D["Summaries and plots, Ch 4, 6"]
  Q -->|Explain or compare| O["Next slide"]
  Q -->|Predict new cases| P["Machine learning, Ch 11–15"]
  Q -->|Find structure| U["PCA, factors, clusters, Ch 9, 14"]
  Q -->|Forecast over time| T["Time series, Ch 16"]

Choosing a method: explaining or comparing

flowchart LR
  O["Outcome type?"] -->|Numeric| N["Repeated or nested data?"]
  O -->|Yes / No| L["Chi-square, logistic regression, Ch 7–8"]
  N -->|No| R["t-test, ANOVA, regression, Ch 7–8"]
  N -->|Yes| M["Mixed-effects models, Ch 10"]

The study’s questions and methods

The study’s questions and methods (continued)

Challenges every researcher meets: data

Challenges every researcher meets: analysis

Where to go next

You do not have to learn alone

• free books: R for Data Science, Regression and Other Stories, ISLR, Tidy Modeling with R, Forecasting: Principles and Practice;
• Posit Community and Stack Overflow for questions;
• R-Ladies and local R user groups;
• TidyTuesday: a new dataset every week to practise on.

A clear question with a small reproducible example is a skill, and the same one that makes an AI assistant useful.

In your field: nutrition science

tooth <- ToothGrowth |> mutate(dose = factor(dose))
tooth |> summarise(mean = mean(len), sd = sd(len), n = n(),
                   .by = c(supp, dose))
summary(aov(len ~ supp * dose, data = tooth))

Prepare, describe, model, report: a complete mini-project in a few lines.

Tooth length depends on dose, form, and their combination: orange juice helps more at low doses, not at the highest.

Practical lab: the Chapter 19 project

Download the complete thesis project from the playground.

Run the cleaning and analysis scripts, then render results.qmd.

Practical exercises 1–3: the pipeline

1. Change one cleaning rule, render again, and note which numbers change.
2. Add a row to Table 1 for students with children.
3. Write a reporting sentence with inline numbers for the effect of support on GPA.

Practical exercises 4–6: figures, methods, reasoning

4. Add a third panel to the thesis figure: wellbeing by faculty.
5. Choose a method for a question from your own field.
6. Complete the chain of reasoning for RQ2, sleep below 7 hours.

Try this yourself

For your own thesis:

• set up the project structure today, with a README;
• move the raw data into data-raw/ and make it read-only;
• write the chain of reasoning for your main research question;
• decide which reporting standard applies.

Troubleshooting guide (Part 1)

Troubleshooting guide (Part 2)

Completion checklist

• the project runs from the raw export to the last table;
• raw data is read-only, scripts numbered, paths relative;
• the cleaning script ends with checks;
• the sample is described in Table 1;
• every result has an effect size and interval;
• every number in the text comes from code;
• each conclusion matches its design;
• limitations are stated, and a reporting standard followed.

Misconceptions to leave behind (Part 1)

Misconceptions to leave behind (Part 2)

The chapter in one sentence

Build the thesis as one project that runs from the raw data to every number in the text, and keep each result attached to the design and limits that give it meaning.

The whole course in one picture

flowchart TB
  subgraph R1[" "]
    direction LR
    A["A question worth answering"] --> B["Clean, checked data"] --> C["The simplest fitting method"]
  end
  subgraph R2[" "]
    direction LR
    D["Results with size and uncertainty"] --> E["Honest conclusions and limits"]
  end
  R1 --> R2

From data to thesis: every step written in code, every claim traceable.

Questions

What is the first thing you will change in your own project?

Which link in your chain of reasoning is weakest?