Glossary

Plain-English definitions of the terms used in these two days.

R and programming

Object: anything with a name in R. A number, a dataset, a plot, a model.

Vector: a sequence of values of the same type. A column of your data.

Data frame / tibble: a rectangle. Rows are observations, columns are variables. A tibble is a data frame with better printing.

Function: a verb. Takes arguments in brackets, returns something.

Argument: an input to a function. round(x, digits = 2) has two.

Package: a bundle of functions someone else wrote. Install once, load each session with library().

Pipe (|>): passes the result on the left into the function on the right. Makes multi-step code read in the order it happens.

Chunk: a block of code inside a Quarto document, fenced by ```{r}.

Working directory: the folder R considers “here”. Set by opening a project; never by setwd().

Project (.Rproj): a folder RStudio treats as the root of the world.

Environment: the objects that currently exist in memory. Emptied by restarting R, which is a feature.

NA: a missing value. Not zero, not empty text. Arithmetic on NA gives NA, deliberately.

Factor: a categorical variable with a fixed set of levels and an order you can control.

Reproducibility: same data + same code → same result.

Replicability: new data, same method → same conclusion.

Generalisability: the finding holds beyond the setting studied.

Literate programming: code and prose in one document, so the report is generated by the analysis rather than describing it.

Dynamic document: a document whose numbers are computed at render time.

Version control (git): a record of every change, who made it and why.

Provenance: the traceable path from raw data to published number.

YAML: the metadata block between --- fences at the top of a .qmd.

Markdown: plain-text formatting. **bold**, ## heading.

Render: run the code and produce the output document.

Chunk option: a #| line controlling one chunk’s behaviour.

Inline code: `r expression` inside prose. How numbers stay correct.

Cross-reference: @fig-gain, automatically numbered and linked.

CSL: Citation Style Language, the file that formats your references.

Shortcode: a {{< >}} directive expanded at render time.

Freeze: caching, so unchanged chunks are not re-run on every render.

Confounding: a third variable driving both exposure and outcome.

Baseline comparability: whether the groups were similar before the intervention. Check it before interpreting any difference after.

Complete-case analysis: analysing only rows with no missing values. What na.rm = TRUE silently gives you.

Effect size: how big the difference is. A p-value does not tell you.

Multiple comparisons: testing many hypotheses inflates false positives.

Pre-specification: deciding the analysis before seeing the results.

Large language model (LLM): a model that predicts the next token. Optimised for plausibility, not truth.

Prompt: what you type. Context + task + constraints + format.

Hallucination: fluent, confident, false output. Citations and function arguments are the usual victims.

Token: the unit an LLM reads and writes; roughly a word fragment.

Disclosure: the statement in your manuscript naming which tool did what. Required by ICMJE.