A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Systematic observation records behavior in natural or structured settings using predefined coding schemes. Codes operationalize constructs (e.g., 'aggression' = hitting, yelling, insults); observation is systematic (e.g., continuous, time-sampled). Multiple coders assess inter-rater reliability; codes are validated against criterion measures. Observation captures behavior directly without self-report bias.
Design a coding scheme for a behavior of interest, specify anchor points for each code, and code a video sample. Compare your codes with a colleague to check reliability. Discuss observational biases (observer effects, selective attention) and methods to minimize them.
Once you have an operational definition of a variable — a precise specification of what you will measure — the question becomes *how* to actually capture that variable in the real stream of behavior. Self-report asks people to characterize their own behavior from memory; systematic observation instead records behavior *as it occurs*, using a predefined system for translating behavioral events into data. The operational definition is expressed as a coding scheme: a set of categories with explicit anchor points that specify exactly what counts as an instance of each behavioral code.
Consider studying aggression in preschool children. Your operational definition might specify: "physical aggression = any intentional act aimed at causing physical harm, including hitting, kicking, biting, and throwing objects at a person." The coding scheme translates this into behavioral markers that a trained observer can reliably identify from video footage. The scheme must be specific enough that two independent observers watching the same footage arrive at the same categorization. That agreement is measured as inter-rater reliability, typically using Cohen's kappa (which corrects for chance agreement) or intraclass correlation coefficients. High kappa confirms that your categories are clear and unambiguous enough to be applied consistently; low kappa signals that coders are making different interpretive decisions, and the scheme needs revision — clearer definitions, worked examples, or recalibration sessions.
The observation method itself shapes the data. Continuous recording captures every instance of a target behavior across a session — appropriate when individual events are discrete and their frequency matters. Time-sampling divides the session into fixed intervals (say, 10-second windows) and records whether the behavior occurred during each interval — appropriate when behaviors are too frequent or too continuous to count individually, or when the goal is estimating the proportion of time spent in a behavioral state. These methods produce different data structures: frequency counts versus proportions, with different implications for statistical analysis.
What distinguishes systematic observation from casual watching is the explicit standardization of inference. Naively, observation seems like pure description — "I just wrote down what I saw." But every behavioral category involves interpretation: is that shove playful or aggressive? Is that sustained gaze attentive or challenging? The coding scheme makes those interpretive decisions in advance, explicitly and consistently, before any data are collected. This is what makes observation scientific: not the absence of judgment, but the disciplined standardization of judgment. When inter-rater reliability is high, it means the standardization has succeeded — independent observers are making the same interpretive decisions, which means the data carries the same meaning across coders, sessions, and sites.