HookLabov got his data by asking for the fourth floor
In 1962, William Labov walked into three New York department stores — Saks Fifth Avenue at the top of the market, Macy's in the middle, S. Klein at the bottom — and asked shop assistants where to find departments he already knew were on the fourth floor. 'Fourth floor': two chances to pronounce an /r/ after a vowel, the sound New Yorkers had come to associate with speaking 'properly'. Then he leaned in — 'sorry?' — and got the phrase again, careful this time. He walked away and noted it down. Two hundred and sixty-four unwitting informants, roughly six and a half hours of fieldwork, and one of the most famous findings in linguistics: the smarter the store, the more the /r/ — with Macy's staff, mid-market and aspirational, shifting hardest when they repeated themselves.
Nothing about that study was expensive. What made it brilliant was design: a question sharp enough to force the variable to show itself twice, casually and then carefully, inside four syllables. Your NEA investigation is the same game at A-level scale — 2,000 words on a question you chose, answered with data you collected yourself. AO1 pays for the methodology, the analysis and the academic expression; AO2 for the concepts and research your study plugs into; AO3 for reading context in every line of your data. The design decisions you make in week one settle most of those marks before a word of the write-up exists.
ModelFrom topic to research question — sharpening the spade
A topic is a field; a research question is the spade that digs it. Watch one sharpen. 'Language and gender' is a shelf of books, not a study. 'How do commentators talk about men's and women's football?' is better — a real data source exists. 'In matched five-minute segments of televised commentary, do commentators use more effort-based than skill-based evaluatives for women players, and first names more often?' is an investigation: countable features, comparable data, a claim the evidence could actually refute.
Each turn of the screw adds something markable — a defined data set, an operationalised feature, falsifiability. Your rationale then starts the AO2 work on page one: why this is linguistically interesting, and which existing debate your study walks into. Two research questions is plenty; three is the ceiling. Never frame one as 'prove that…' — an investigation built to prove something has written its conclusion before collecting its evidence, and the confirmation bias leaks into every analytical paragraph. Finally, apply the feasibility test: can you genuinely obtain this data within a few weeks, with consent where needed? The perfect corpus you never collected scores nothing.
ModelMethodology — data you can defend
Comparability is the spine of method marks: hold genre, situation and length constant, and vary only the thing under test. Same broadcaster, same competition, same match phase — then any gap you find has a fighting chance of meaning something. From there, three collection routes dominate. Record and transcribe spoken interaction (consent first, always). Assemble a corpus of written or electronic texts, matched by text type and date — front pages, gravestone inscriptions, WhatsApp openings, estate-agent listings. Or elicit: questionnaires and attitude-rating tasks in the tradition of matched-guise work.
Be honest about size. At 2,000 words, small-and-controlled beats big-and-mushy: ten minutes of talk transcribed properly and analysed at three language levels outworks an hour skimmed at one. Transcribe with declared conventions and a key — pauses, overlaps, emphasis — because the transcript is evidence, and evidence needs a chain of custody. Ethics is not paperwork: informed consent for recordings, anonymisation in the write-up, parental consent for children's data, and a real distinction between broadcast or public text (fair game) and the family group chat (everyone's yes, or nothing). Then pilot the method on a fragment before committing your weekends — Labov's design survived contact with reality because it was tested cheap.
MechanismAnalysis — quantify, quote, explain
The analysis is the biggest room in the house — well over half your words — and it runs on a three-stroke loop. Quantify the pattern: counts, normalised per 1,000 words when your texts differ in size. Quote the sharpest instance. Explain it in context: who is speaking, to whom, in what mode, with what at stake. That third stroke is AO3, and it belongs inside every analytical paragraph, not in a bolted-on 'context' section at the end. The standing complaint of moderators is feature-spotting — 'here is another interrogative' — description that never becomes interpretation. The cure is letting the levels interlock: a lexical pattern doing pragmatic work, a syntactic habit serving a discourse purpose.
Suppose your matched five-minute commentary segments return these counts. Men's match: 14 skill-frame evaluatives ('unplayable', 'brilliant recovery'), 3 effort-frame; surname-only address throughout. Women's match: 6 skill-frame, 11 effort-frame ('worked so hard', 'never stops running'), and first-name address nine times. A top-band paragraph runs the loop: 'Effort-frame evaluatives outnumber skill-frame nearly two to one in the women's segment (11:6), reversing the men's ratio (3:14) — an asymmetry consistent with the 1990s American television studies that found commentary crediting sportsmen with talent and sportswomen with industry, and first-naming women far more readily. The pragmatics sharpen the pattern: “she's worked so hard for that, bless her” wraps its praise inside an endearment no commentator offers a male midfielder, framing achievement as touching rather than expected. Context, though, complicates the count: both matches are live and co-commentated, and effort-praise clusters around defensive passages in both segments — so match-state, not gender alone, may drive part of the gap.' That final sentence — an alternative explanation offered against your own thesis — is exactly the move that separates analysis from advocacy.
ModelResearch that earns its place — AO2 without the wallpaper
Three to five studies, chosen because they bear on your data, used as lenses rather than furniture. A gender investigation might set Zimmerman and West's 1975 interruption findings against Geoffrey Beattie's much larger samples from the early 1980s, which found interrupting far more evenly spread between the sexes; add O'Barr and Atkins' 1980 courtroom study, which re-read 'women's language' as powerless-in-situation language tracking status rather than sex; and Deborah Cameron's The Myth of Mars and Venus (2007), which argues that variation within each gender swamps the average difference between them. An accent or dialect study reaches for Labov, for Trudgill's 1974 Norwich finding that men claimed more non-standard forms than they actually used — covert prestige in the wild — and for Kerswill and Cheshire's work on Multicultural London English.
The top-band move is arguing with a study: 'my data resists Zimmerman and West because…'. A finding from 1975 is a hypothesis your data gets to test, not a law it must obey — and a conclusion that refines an old claim reads as scholarship, not summary.
MechanismThe write-up — 2,000 words of academic register
The shape moderators expect: aims and rationale (brief), methodology (justify every choice, own the limits), analysis (the majority of the budget), conclusion (answer each research question by name, scale the claims to the sample, point at extensions), references in one consistent convention, and appendices carrying the raw data, transcripts with their key, and consent evidence — outside the running word count. Quote data sparingly and precisely; the appendix holds the haystack so the essay can hold the needles.
Academic register is precision, not pomposity: exact verbs, past tense for what you did, present tense for what the data shows. Keep a research log from day one — methods considered, dead ends hit — and the limitations paragraph writes itself, honestly. Honesty about scope is a strength: 'this holds for two commentators on one broadcaster' is a defensible claim; 'commentators are sexist' is not. Remember who you are writing for: the folder is marked by your teacher and moderated by AQA, so it has to persuade a stranger who was never in the room. Treat 2,000 as a hard ceiling — the discipline itself reads as control.
VocabularyKey terms the mark scheme pays for
TrapsMisconceptions that cost marks
ExamWhat examiners want
The NEA folder is worth 20% of the A-level and the investigation is one of its two halves — marked by your teacher against AQA's criteria and moderated by the board, so every page must persuade a stranger who was never in the room.
Hit each assessment objective where it lives. AO1: a methodology that justifies every choice, terminology applied precisely to your own data, and academic expression with one consistent referencing convention throughout. AO2: concepts and named research framing the aims and woven through the analysis as lenses, not dropped in as wallpaper. AO3: contextual reasoning attached to every interpretive claim — and the alternative explanation offered against your own thesis is the single most reliable top-band signal in the component.
Rituals that pay: research questions stated on page one and answered by name in the conclusion; the quantify–quote–explain loop in every analytical paragraph; appendices for everything raw, with a transcription key; a research log from day one so the limitations section is honest rather than defensive. And choose data you can genuinely get. The investigation rewards the student who designs like Labov — cheap, sharp and repeatable — not the one with the grandest unobtainable corpus.