Written and maintained by CASRAI Editorial Board
Last updated
Naturalistic observation is a qualitative data-collection method in which a researcher watches and records behaviour as it occurs in its normal setting, without manipulating any variable or controlling the environment the way an experiment does. The defining commitment is ecological validity: behaviour recorded in a classroom, a workplace, a public street or an animal’s natural habitat is behaviour the researcher did not stage.
This is a field-observation method, part of the qualitative-methods toolkit alongside interviews and document analysis — not the same concept as an observational study design in the epidemiological sense (cohort, case-control, cross-sectional studies that observe outcomes without assigning treatment). Both traditions use the word “observational,” but they answer different questions with different tools; this guide covers the field-method sense used in psychology, education, sociology, organisational research and ethology.
What makes a study “naturalistic observation”
Three features distinguish naturalistic observation from other data-collection methods:
- Real setting. Data is gathered where the behaviour normally happens — a playground, a hospital ward, a customer-service call, a forest — not a lab, survey instrument or interview room.
- No experimental manipulation. The researcher does not introduce a stimulus, assign conditions or control extraneous variables. This is what separates it from a field experiment, which also happens in a natural setting but does manipulate something.
- Minimal researcher interference. The observer aims to record what would have happened anyway, as closely as the observation method allows. In practice this is a matter of degree, not an absolute — see observer effects below.
It sits alongside the other core qualitative data-collection methods covered in the qualitative research methods guide and produces the kind of unstructured, context-rich material discussed in what qualitative data is. It is also frequently used as one data source within a broader case study design, alongside interviews and documents.
Covert vs. overt observation
The first design decision is whether the people being observed know a researcher is watching.
- Overt observation: participants know they are being observed and, typically, have given informed consent. This is the ethically straightforward default and is required whenever the setting is not genuinely public or the behaviour is sensitive.
- Covert observation: the researcher observes without participants’ knowledge, usually to avoid reactivity (people behaving differently because they know they’re watched — see the observer effect below). Covert observation is ethically permissible only under narrow conditions: the setting must be genuinely public, no personal identification or harm can reasonably result, and there is no feasible way to answer the research question with consent obtained in advance. The American Psychological Association’s Ethical Principles of Psychologists and Code of Conduct (Standard 8.05, “Dispensing With Informed Consent for Research”) allows consent to be dispensed with specifically for naturalistic observation in public places where disclosure would not place participants at risk, and a related recording provision (Standard 8.03) permits recording without consent for research consisting solely of naturalistic observation in public places where identification or harm is not anticipated. Institutional ethics/IRB approval is still required to make that judgment — a researcher does not get to decide unilaterally that an observation is low-risk enough to skip review.
Covert observation trades reactivity for a real cost: participants cannot object, correct a misreading of their behaviour, or withdraw. Most covert naturalistic observation is restricted to genuinely public behaviour (how pedestrians use a crosswalk, how shoppers queue) rather than anything that could identify or expose an individual.
Structured vs. unstructured recording
The second design decision is how tightly the recording itself is predefined.
- Unstructured recording (field notes): the observer writes a running, open-ended description of what happens, without a predetermined category system. This produces rich, contextual data well suited to exploratory questions and to grounded, inductive analysis, but is slower to analyse and more exposed to observer bias in what gets noticed and written down.
- Structured recording (coding schemes, checklists, rating scales): the observer defines behavioural categories in advance and records only whether/how often/how long each occurs. This produces data that can be quantified and compared across observers or sessions, at the cost of missing anything outside the predefined categories.
Many naturalistic observation studies combine both: structured tallying against a coding scheme for the behaviours the study is specifically testing, plus unstructured field notes to capture unanticipated context.
Sampling behaviour: how much of the stream you record
Because continuous behaviour cannot always be recorded in full, naturalistic observation studies choose a sampling strategy for deciding which moments of behaviour to capture:
| Sampling method | How it works | Best suited to |
|---|---|---|
| Event sampling | Record every instance of a predefined target behaviour whenever it occurs, for the full observation period. | Behaviours that are relatively infrequent or discrete (e.g. every instance of a specific classroom disruption). |
| Time sampling | Observe and record only during predetermined time windows (e.g. the first 30 seconds of every 5-minute interval), rather than continuously. | Frequent or continuous behaviours where recording everything is impractical; reduces observer fatigue. |
| Interval sampling | Divide the observation period into fixed intervals and record only whether the behaviour occurred at all within each interval (partial-interval) or throughout it (whole-interval). | Studies needing an estimate of behaviour prevalence rather than precise counts or durations. |
| Instantaneous / point-time (scan) sampling | Record whether the behaviour is occurring at a single predetermined instant (a “scan”), often across multiple subjects in turn. | Group or multi-subject settings (a classroom, an animal group) where one observer must sample several individuals. |
| Continuous / narrative recording | Record the full, uninterrupted stream of behaviour and context for the observation period. | Short observation periods, or exploratory work where the categories of interest aren’t yet known. |
Observer effects and reactivity
Any time a participant’s awareness of being watched changes the behaviour being measured, the resulting distortion is a threat to the study’s validity. This is documented in more depth in the observer effect guide and its best-known specific case, the Hawthorne effect (workers changing behaviour specifically because they know they are part of a study). Naturalistic observation studies manage this risk primarily through:
- Habituation: observing over a long enough period that participants stop attending to the observer’s presence.
- Unobtrusive positioning: observing from a distance, through one-way glass, or via already-present recording equipment rather than a visibly note-taking researcher.
- Covert observation, where ethically permissible (see above) — the strongest control against reactivity, but the most ethically constrained option.
Inter-observer reliability
Because naturalistic observation depends on a human observer’s judgment about what counts as an instance of a behaviour, a second, independent observer coding the same session is the standard check that the coding scheme — not one observer’s idiosyncratic reading — is producing the data. Agreement between observers is generally reported as percentage agreement for simple counts, or more rigorously as an intraclass correlation coefficient (ICC) or Cohen’s kappa for categorical/continuous ratings, which correct for the level of agreement expected by chance alone. See reliability in research measurement for how this fits into the broader reliability picture, and test-retest vs. inter-rater reliability for how inter-observer agreement compares to the other main reliability check. Low inter-observer agreement is usually a sign the coding scheme’s category definitions are too vague or too subjective, not that the observers are careless — the fix is almost always to tighten the operational definitions and retrain, then re-test agreement.
Strengths and limitations
| Naturalistic observation | |
|---|---|
| Strengths | High ecological validity (behaviour is not staged); can capture behaviour that would not occur, or would occur differently, under lab conditions; useful where experimental manipulation would be unethical or impractical; can generate hypotheses for later controlled testing. |
| Limitations | No control over extraneous variables, so causal claims are not supportable from observation alone; findings can be time- and context-specific and may not generalise; observer bias and reactivity are real threats; behaviours of interest may occur too rarely to capture in a feasible observation window; covert forms raise real consent and privacy constraints. |
Naturalistic vs. controlled observation
Naturalistic observation is often contrasted directly with controlled observation, where the researcher structures the environment — a standardised task in a lab, a one-way-mirror playroom, a set procedure participants follow — while still recording behaviour rather than self-report. Controlled observation trades ecological validity for standardisation and replicability: every participant faces the same conditions, which makes comparison across participants or sessions far cleaner, at the cost of behaviour that may not reflect what the same person would do in an everyday setting.
Naturalistic observation vs. participant observation
Both are field-observation methods, and the terms are sometimes used loosely as if interchangeable, but they differ on the researcher’s role. In naturalistic observation, the researcher stays outside the activity being studied — a detached observer recording behaviour without joining in. In participant observation (the core technique of ethnography), the researcher takes part in the setting’s ordinary activities alongside the people being studied, trading some observational distance for the interpretive access that comes from lived participation. Naturalistic observation is generally better suited to discrete, codeable behaviours studied from outside; participant observation is generally better suited to understanding meaning, culture and social process from within.
Worked example
A researcher studying turn-taking in toddler play sets up in the corner of a nursery classroom the children already know, observes during normal free-play time (no staged activity), and uses a structured coding sheet with 30-second time-sampling intervals to tally instances of toy-sharing, conflict and adult intervention, while also jotting brief unstructured notes on anything unusual. A second observer independently codes video of the same sessions and the two coders’ tallies are compared for inter-observer agreement before the data is analysed. This combines overt observation (staff and, where age-appropriate, parents were informed and consented), structured time-sampling, and an inter-observer reliability check — the standard shape of a rigorous naturalistic observation study.
Ethics and regulatory status: when observation counts as human-subjects research
Observation is the research method where the ethics question is easiest to get wrong, because “they were in public” feels like it should settle it and does not. Under the US Common Rule (45 CFR part 46), which governs federally supported human-subjects research and which most US institutions apply to all their research by policy, the analysis turns on two definitions and one exemption category. Researchers outside the US should read this as the structure of the question rather than the applicable law, and check their own framework — but the distinctions it draws are the ones every framework has to draw somehow.
Are you studying “human subjects” at all?
A human subject is defined at 45 CFR 46.102(e)(1) as a living individual about whom an investigator either (i) obtains information through intervention or interaction with the individual and then uses it, or (ii) obtains, uses, studies, analyses or generates identifiable private information. Pure observation involves no intervention and, if the observer stays out of the interaction, no interaction either — so the whole question collapses onto the second limb, and specifically onto what counts as “private.”
The regulation defines this with unusual precision, and the definition is written almost as though naturalistic observation were the case it had in mind. Under 46.102(e)(4), private information “includes information about behavior that occurs in a context in which an individual can reasonably expect that no observation or recording is taking place.” Note what the test attaches to: not the physical location, but the reasonable expectation in that context. A conversation in a corner of a hospital cafeteria can satisfy it; a person’s conduct in the middle of a railway concourse generally does not. Under 46.102(e)(5), the information is identifiable private information if the subject’s identity is, or may readily be, ascertained by the investigator or associated with the information — which is where video and audio recording changes the analysis, because a recording of a face or a voice usually is identifiable even when the researcher never learns a name.
The exemption for observation of public behaviour
Where the activity does constitute human-subjects research, 45 CFR 46.104(d)(2) sets out an exemption category that covers “observation of public behavior (including visual or auditory recording)” — but only if at least one of three conditions is met:
- (d)(2)(i) — the information is recorded in such a manner that the identity of the subjects cannot readily be ascertained, directly or through linked identifiers. In practice this is the coding-sheet-only route: tallies and durations against a scheme, no names, no retained recordings.
- (d)(2)(ii) — any disclosure of the subjects’ responses outside the research would not reasonably place them at risk of criminal or civil liability, or be damaging to their financial standing, employability, educational advancement, or reputation. This is the route that fails for observational studies of anything stigmatised, unlawful, or workplace-sensitive, however public the setting.
- (d)(2)(iii) — the information is recorded identifiably, and an IRB conducts a limited IRB review to make the determination required by 46.111(a)(7) (adequate provisions to protect privacy and maintain confidentiality). This is the route most video-based observation studies actually take.
Two practical points follow. First, the exemption determination is an institutional one: the question of whether a study fits a category is normally answered by the IRB or its designee under local policy, not self-certified by the investigator, so the sequence is submit-then-observe rather than the reverse. Second, an exemption is not an ethics waiver. A study can be exempt from the regulation and still require a data-management plan, a retention-and-destruction schedule for recordings, and a defensible answer to the question of what the observed people would say if told afterwards.
The rules are stricter when the observed people are children
This matters directly for the classroom, playground and nursery settings where naturalistic observation is most commonly taught and used. Research involving children is subject to subpart D of the regulation, and 45 CFR 46.104(b)(3) narrows the exemption accordingly: paragraphs (d)(2)(i) and (ii) apply to such research “involving educational tests or the observation of public behavior when the investigator(s) do not participate in the activities being observed,” and paragraph (d)(2)(iii) — the limited-IRB-review route for identifiable recordings — may not be applied to research subject to subpart D at all.
Read together, those two clauses have a sharp consequence for study design. An observer who joins the activity — helping with the game, prompting the group, taking a role in the classroom — has stepped outside the exemption entirely, which is one more reason the participant/non-participant distinction covered above is a design decision rather than a stylistic one. And a study of children that keeps identifiable video cannot reach the exemption by the third route; it needs review through the ordinary channel, with parental permission and, where age-appropriate, child assent handled under subpart D.
Covert observation is where the analysis actually bites
The covert designs discussed earlier in this guide are precisely the ones that stress these tests, because the justification for not telling people is usually that telling them would change the behaviour being studied — which is a scientific rationale, not a regulatory one. The regulation does not contain an exception for “consent would have spoiled the data.” Covert observation is defensible when the behaviour genuinely occurs in a context carrying no reasonable expectation of privacy, and when the record is kept in a form that satisfies one of the three conditions above; it is not defensible merely because the data would otherwise be reactive. Where an ethics committee does approve covert work in a semi-private setting, it will normally attach conditions — restricted recording, rapid de-identification, defined destruction, and often debriefing — and those conditions are part of the method, not paperwork attached to it.
Frequently asked questions
Is naturalistic observation qualitative or quantitative?
It can produce either, or both. Unstructured field notes are qualitative data. Structured recording against a predefined coding scheme (tallies, durations, frequencies) produces quantitative data that can be statistically analysed. Many studies do both in the same observation session.
What is the difference between naturalistic observation and a case study?
Naturalistic observation is a data-collection method. A case study is a broader research design that can incorporate naturalistic observation as one of several data sources (alongside interviews, documents, and archival records) focused on a single case, setting or instance.
Can naturalistic observation establish cause and effect?
No. Without manipulating a variable or controlling for confounds, naturalistic observation can describe and correlate behaviour but cannot support causal claims on its own. It is often used to generate hypotheses that a controlled experiment then tests.
What are examples of naturalistic observation?
Watching how commuters navigate a train platform without staging the situation, recording peer interactions during unstructured school recess, observing animal foraging behaviour in the wild, or coding customer behaviour on a retail sales floor from an unobtrusive vantage point are all naturalistic observation — the common thread is a real setting with no manipulated variable.
Do researchers need consent to use naturalistic observation?
Usually yes, and overt, consented observation is the default. Consent can be dispensed with only under narrow, ethics-board-approved conditions: the setting is genuinely public, no personal identification or harm is anticipated, and the research question could not reasonably be answered with prior consent. See the covert vs. overt section above.
What is naturalistic observation in psychology?
The same method described on this page, applied to human behaviour in psychological research: watching and recording behaviour in the setting where it normally occurs, with no variable manipulated and no task assigned. Psychology uses it most heavily in developmental research (children in classrooms and playgrounds), social psychology (interaction in public settings), and clinical research (behaviour in wards or homes), typically as a source of hypotheses that a controlled experiment then tests, or as an ecological-validity check on findings first obtained in a laboratory. The word “naturalistic” refers to the setting being the participant’s own, not to the data being qualitative — structured coding in a natural setting is still naturalistic observation, and yields quantitative data.
Which is the best example of naturalistic observation?
The strongest examples share four features rather than a particular subject matter: the setting is one the observed people or animals would have been in anyway, no variable is manipulated, the recording follows a scheme fixed before observation began, and a second observer independently codes a portion of the material so inter-observer agreement can be reported. Long-term field studies of animal behaviour in the wild are the archetype because they satisfy all four over years. A nursery-classroom study using 30-second time sampling and a second coder, like the worked example above, satisfies all four in a single term. A researcher informally watching a setting and writing impressions afterwards satisfies none of them and is not a study.
Is naturalistic observation research exempt from IRB review?
Sometimes, but that is a determination the IRB makes, not the researcher. Under the US Common Rule, observation of public behaviour can fall within the exemption at 45 CFR 46.104(d)(2) if the record is not identifiable, or if disclosure would not put subjects at risk of liability or damage to reputation, employability, financial standing or educational advancement, or if an IRB conducts a limited review of an identifiable record. Studies involving children face tighter limits, and covert observation in settings carrying a reasonable expectation of privacy generally falls outside the exemption entirely. See the ethics and regulatory section above for the full analysis.
Is naturalistic observation used outside psychology?
Extensively. Ethology and behavioural ecology are built on it. Anthropology and sociology use it as the observational component of fieldwork, where it shades into participant observation as the researcher takes on a role in the setting. Health-services and implementation research use it to record what clinical teams actually do rather than what protocols say they do, and human-factors research uses it to study how people really operate equipment in situ. The method is the same across all of them; what changes is how structured the recording is and how far the observer participates.
