You are viewing a free preview of this lesson.
Subscribe to unlock all 10 lessons in this course and every other course on LearningBro.
Knowing that observations and self-reports exist (Lesson 1) is not the same as being able to design one that produces trustworthy data. This lesson is about the craft: how to turn "watch children playing" into a system that two observers could apply identically, and how to turn "ask people about stress" into questions that measure what you intend without leading the respondent. OCR examines this directly — "design a coding frame for this observation" and "write a suitable closed question" are recurring Component 01 tasks, and you build these tools yourself in your Practical Activities.
We cover the machinery of observation design — behavioural categories, coding frames, and the two ways of sampling behaviour over time: time sampling and event sampling. We then move to self-report design — open versus closed questions, rating scales (Likert and semantic differential), and the principles of writing valid, unbiased items free of leading, ambiguous or double-barrelled wording.
| This lesson covers | OCR H567 Component 01 sub-area | AO focus |
|---|---|---|
| Behavioural categories and coding frames | 1.2 Planning & conducting — designing observations | AO1; AO2 designing a coding frame |
| Time sampling and event sampling | 1.2 — designing observations | AO1; AO2; AO3 |
| Open vs closed questions | 1.2 — designing self-reports | AO1; AO2 |
| Rating scales: Likert, semantic differential | 1.2 — rating scales | AO1; AO2 |
| Writing valid, unbiased items | 1.2 — designing self-reports; links to validity | AO2; AO3 |
Referenced descriptively; see the official OCR H567 specification document for exact wording. This lesson develops AO1 (defining the design tools), AO2 (constructing a coding frame or writing a question for a scenario — a common exam task) and AO3 (evaluating design choices for reliability and validity).
The strength of an observation depends almost entirely on the quality of its recording system. Two observers watching the same behaviour must agree — otherwise the data is unreliable.
Recall from Lesson 1 that an observation can be structured (using predefined categories) or unstructured (recording rich detail freely), and this choice frames everything that follows. A structured observation front-loads the design work: the researcher decides in advance exactly which behaviours to count, builds a coding frame, and produces quantitative data that is easy to compare and to test statistically — at the cost of only capturing what the frame anticipated. An unstructured observation defers that structure: the observer records as much as possible in qualitative notes, which is invaluable when the researcher does not yet know which behaviours will prove important (an exploratory study of a novel setting), but yields data that is hard to quantify, harder to compare, and more exposed to the observer's selective attention. In practice, unstructured observation often precedes structured observation: an exploratory phase reveals the behaviours that matter, which are then operationalised into categories for a later structured study. The two are stages in a research programme as much as rival techniques.
A behavioural category is a specific, observable, operationalised action that "counts" as an instance of the behaviour being studied. The art is to break a broad concept ("aggression", "engagement", "attachment behaviour") into a set of discrete, mutually-exclusive, objectively-recordable categories.
Good behavioural categories are:
Each of these properties exists to serve the same master: the elimination of observer disagreement. A frame that meets all four turns the inherently subjective act of "watching behaviour" into something approaching an objective measurement, because it removes the judgement calls that would otherwise let two observers see the same event differently. This is why the effort invested in designing categories before observation begins is repaid many times over in the quality and defensibility of the resulting data.
A coding frame (or behaviour schedule) is the complete list of behavioural categories, together with the symbols or tally system used to record them, that observers use during the observation. It is the observation's equivalent of a questionnaire — the instrument that standardises data collection. A well-designed coding frame is what makes inter-rater reliability achievable.
The design of a coding frame is where an observation succeeds or fails, and the reasoning repays close attention. The central problem an observation must solve is subjectivity: two people watching the same scene can genuinely disagree about what they saw, because "aggression" or "engagement" are interpretive labels. A coding frame tames this by forcing the researcher to decide, in advance and in concrete terms, exactly what will count. The more the categories describe observable actions ("pushes another child") rather than inferred states ("is being hostile"), the less room there is for two observers to diverge, and the higher the inter-rater reliability. This is why the criteria for good categories — operationalised, objective, mutually exclusive, exhaustive — are not arbitrary style rules but the very features that make observational data trustworthy. A poorly-designed frame with overlapping or vague categories produces data that reflects the observers' idiosyncratic judgements more than the participants' behaviour, and no amount of careful watching can repair it after the fact.
A practical tension arises between having too few and too many categories. Too few, and the frame is crude — it lumps distinct behaviours together and loses information. Too many, and the observer cannot realistically track them all in real time, especially if behaviour is fast-moving, so reliability falls as the observer struggles to keep up. The art is to choose a manageable set of well-defined categories that capture the behaviours that matter for the research question, plus an "other" category to catch the unanticipated. Piloting the frame — trialling it on a sample of behaviour and refining categories that prove ambiguous or unused — is standard good practice before the main observation.
| Behavioural category (operationalised) | Tally symbol |
|---|---|
| Physical aggression — hits, kicks or pushes another child | P |
| Verbal aggression — shouts, name-calls or threatens | V |
| Cooperative play — shares a toy or takes turns | C |
| Solitary play — plays alone with no interaction | S |
| Other observable behaviour | O |
Because you usually cannot record everything continuously, you must decide when to record. OCR names two strategies.
Time sampling records which behaviours occur within fixed time intervals — for example, noting the child's activity in a 10-second window every minute. It makes data collection manageable and produces comparable snapshots, but it may miss behaviours that happen outside the sampled intervals, and rare-but-important events can be overlooked.
Event sampling records every occurrence of a specified behaviour throughout the observation, regardless of when it happens — for example, tallying every instance of physical aggression across the whole session. It captures complete frequency data for the target behaviour and does not miss events, but it can overwhelm a single observer if events happen rapidly or several occur at once.
The choice between time and event sampling is a genuine trade-off between feasibility and completeness, and matching the method to the behaviour is the examinable skill. Time sampling is the sensible choice when the target behaviour is frequent or more or less continuous — if a child is almost always doing something codeable, trying to record every act is impossible, so taking snapshots at fixed intervals makes the task manageable and yields comparable data points across children or sessions. Its price is that behaviour occurring between the sampled moments goes unrecorded, so a rare but important act can be missed entirely. Event sampling is the sensible choice when the target behaviour is discrete and countable and especially when it is infrequent — you want a complete count of, say, every act of helping, and because such acts are occasional a single observer can catch them all without being swamped. Its price is that when events cluster or occur simultaneously, one observer cannot keep up and data is lost. A sophisticated answer does not merely define the two methods but justifies a choice for a given behaviour: "because aggressive incidents are relatively infrequent and discrete, event sampling will capture a complete and accurate frequency, whereas time sampling might miss incidents falling between intervals."
Both methods, incidentally, can be combined with either a structured or unstructured approach and with any point on the participation and awareness dimensions from Lesson 1 — sampling in time is a separate design decision from the four dimensions of type. A single observation must be specified on all of these choices.
Designing an observation also forces an ethical decision that self-reports usually avoid: whether the observation will be overt (participants know) or covert (they do not). Covert observation is methodologically attractive because it eliminates the reactivity that awareness produces — people who do not know they are watched behave naturally, free of demand characteristics and social desirability. But it carries obvious ethical costs: participants cannot give informed consent, and covert observation of people in situations where they would expect privacy is a serious breach. The usual resolution is that covert observation is more defensible in public settings where behaviour is already on view (a busy street, a shopping centre) and where individuals are not identifiable, and far less defensible in private ones. The design of an observation therefore cannot be separated from the ethics of it — a point examined further in Lesson 9 — and a strong design answer flags the consent and privacy implications of a covert choice rather than treating the decision as purely technical.
| Time sampling | Event sampling | |
|---|---|---|
| What is recorded | Behaviour within fixed intervals | Every occurrence of a target behaviour |
| Strength | Manageable; comparable snapshots | Complete frequency; misses nothing |
| Weakness | May miss behaviour between intervals | Can overwhelm observer if events are frequent |
| Best when | Behaviour is frequent/continuous | Target behaviour is discrete and countable |
graph TD
A["Designing an observation"] --> B["Define behavioural categories<br/>(operationalised, objective, exclusive)"]
B --> C["Assemble a coding frame<br/>(categories + tally system)"]
C --> D{"When do we record?"}
D -->|"At fixed time intervals"| E["Time sampling<br/>manageable, but may miss events"]
D -->|"Every time the behaviour occurs"| F["Event sampling<br/>complete count, but can overwhelm observer"]
C --> G["Two observers code independently<br/>→ check inter-rater reliability"]
style E fill:#2980b9,color:#fff
style F fill:#27ae60,color:#fff
style G fill:#8e44ad,color:#fff
A self-report is only as good as its questions. Poor wording produces data that measures the question's flaws rather than the participant's true state.
A closed question offers a fixed set of response options (yes/no, multiple choice, a rating scale). It yields quantitative data that is quick to analyse and compare across many respondents, but it constrains answers to the options provided and may miss a respondent's real view.
A open question invites a free-text answer in the respondent's own words. It yields rich qualitative data with depth and nuance, but is time-consuming to analyse and harder to compare across respondents.
The decision between open and closed questions is, once again, a trade-off with consequences that ripple through the whole study. Closed questions are efficient and objective: they can be answered quickly, distributed to large samples, and analysed statistically with little interpretation, which is why surveys of attitudes and opinions lean heavily on them. But their fixed options impose the researcher's framework on the respondent — if none of the options fits, the respondent must choose the "least wrong" answer, and the data then misrepresents their real position. Open questions reverse the trade-off: they let respondents express views the researcher never anticipated, capturing nuance and meaning that closed items would flatten, but the resulting free text must be coded or content-analysed before it can be summarised, which is laborious and introduces the researcher's interpretation as a possible source of bias. Many well-designed questionnaires therefore mix the two — closed items for the core measurable variables, a few open items to capture context and unanticipated responses — gaining breadth and depth together, at the cost of a more demanding analysis.
There is also a validity dimension. Closed questions can inadvertently lead respondents by the options they offer or omit, and can miss the true answer entirely; open questions avoid imposing options but are more vulnerable to inconsistent interpretation and to social desirability, since respondents compose their own answers and may curate them. Neither format is intrinsically more valid — the validity depends on how carefully the questions are written and matched to the research aim.
| Closed question | Open question | |
|---|---|---|
| Response | Fixed options | Free text |
| Data | Quantitative | Qualitative |
| Analysis | Quick, comparable | Detailed but laborious |
| Risk | May miss the true view | Hard to summarise/compare |
Rating scales are a common form of closed question that quantify attitudes.
A Likert scale presents a statement and asks respondents how strongly they agree, usually on a five- or seven-point scale from "strongly disagree" to "strongly agree". For example: "I feel confident speaking in front of my class" — Strongly disagree · Disagree · Neither · Agree · Strongly agree. Likert scales produce ordinal data (the points are ordered, but the gaps between them are not guaranteed equal).
Subscribe to continue reading
Get full access to this lesson and all 10 lessons in this course.