All subjects · English Language · WEN04 Investigating Language
Collecting data: methods and sampling
Edexcel IAL English Language · WEN04 Investigating Language
Your Section B argument is only as good as your data. Learn how linguists collect spoken and written data, how to choose a sample, and the main problem with recording people.
The observer's paradox
Labov described the aim of studying 'how people talk when they are not being systematically observed', yet the only way to get that data is to observe them (Labov 1972, as quoted in Essentials of Linguistics). Recording people makes them monitor their speech. Remedies: build rapport, record for longer so people relax, ask questions that prompt stories, and use texts that already exist (broadcasts, posts, published documents).
Methods
| Method | Strength | Limitation |
|---|---|---|
| Sociolinguistic interview: casual talk, reading passage, word list | Collects speech at different levels of self-monitoring | Observer's paradox; time to transcribe |
| Observation and field notes | Natural behaviour in context, e.g. caregivers with children | Hard to note everything; needs consent |
| Rapid anonymous survey | Labov asked New York store staff a question whose answer was 'fourth floor' to hear the /r/; 264 people in two days | Short, single feature; ethics today would need careful review |
| Existing texts (documents, websites, broadcasts) | No observer effect; easy to collect | May be edited or scripted; context must be checked |
| Questionnaires on attitudes | Many responses quickly | Self-reports can be inaccurate (Trudgill's Norwich study found self-evaluations inaccurate) |
From Essentials of Linguistics 10.5, Nautilus on Labov's department store study, and Trudgill (1972).
Sampling
One person shows only one person's patterns. Essentials of Linguistics says a good sample is representative and, ideally, socially stratified: roughly equal numbers from each social group you want to compare (age, class, gender, region). Record the details of every speaker or text: age, setting, audience, date. For written data, choose comparable texts (same genre, similar length) so differences are not just caused by genre.
Transcribing
Transcribe using the Transcription Mark Key conventions you know from Units 1 and 2: (.) micro pause, ? rising intonation, phonemic symbols for key pronunciations. Transcribe accurately and do not 'correct' the speaker.
Worked example
You want to compare how caregivers manage behaviour with 2-year-olds and 6-year-olds. Design a sample.
- Groups: caregivers of 2-year-olds and caregivers of 6-year-olds.
- Stratify: similar numbers in each group, for example four families each.
- Settings: the same situations (tidying up, mealtimes) so the comparison is fair.
- Record details: ages, setting, who is present; reduce the observer's paradox by recording over several days.
Answer: A small, stratified, comparable sample with matched settings, recorded over time to reduce self-monitoring.
Check yourself
- The observer's paradox means that
- observers always mishear speech
- data must be collected by two people
- the act of observing people changes how they speak
- written texts cannot be analysed
Show answer
the act of observing people changes how they speak. People monitor their speech when they know they are recorded.
- A socially stratified sample
- includes roughly equal numbers from each social group being compared
- uses only one speaker
- selects people at random from one group
- includes only written texts
Show answer
includes roughly equal numbers from each social group being compared. Essentials of Linguistics: roughly equal numbers from each relevant group.
- In Labov's department store study, the question was chosen so that the answer was
- 'thank you'
- 'yes'
- 'fourth floor'
- 'I don't know'
Show answer
'fourth floor'. 'Fourth floor' contains two chances to pronounce or drop /r/.
- A key advantage of collecting existing texts such as public web pages is
- they are always unedited
- no observer effect on how they were produced
- they never need checking for context
- they are always spoken
Show answer
no observer effect on how they were produced. They were produced without your involvement, though they may be edited.
- Labov surveyed 264 people over two days. How many is that per day, on average?
Show answer
132 people. 264 / 2 = 132.
- Trudgill's Norwich study warns that attitude questionnaires
- always reflect real speech
- can be inaccurate because self-reports differ from real use
- only work with children
- are banned in linguistics
Show answer
can be inaccurate because self-reports differ from real use. Trudgill found speakers' self-evaluations were inaccurate.
- Give two ways to reduce the observer's paradox when recording speech.
Show answer
Build rapport and record for longer so speakers relax, and ask questions that get them telling stories.. Essentials of Linguistics suggests rapport, good questions and genuine follow-up interest.
Sources
- Pearson Edexcel International Advanced Subsidiary/Advanced Level in English Language specification (Issue 4, February 2026)
- Anderson and others (2022), Essentials of Linguistics, 2nd edition, section 10.5 Variationist methods and concepts (eCampusOntario open textbook)
- Nautilus (2019): Why working-class New Yorkers drop their 'Rs' (on Labov's department store study)
- Trudgill (1972), Sex, covert prestige and linguistic change in the urban British English of Norwich, Language in Society 1(2) (Cambridge University Press)
Last updated 8 November 2026. Spotted a mistake? Tell us.