Identifying a good research question is a vital first step in any behavioural study because the question will focus the rest of the research cycle. Four logically distinct types of question can be asked about any behaviour. These concern its mechanisms, its development (or ontogeny), its function and its evolution (or phylogeny). The mechanisms underlying behaviour can be studied at many different levels, ranging from the social or physical environmental conditions that influence the behaviour down to the neural networks responsible for behavioural output. The nature of the research question will influence decisions about what species to study. Research questions are developed through a combination of approaches, including reading the literature, preliminary observations and exploratory data analysis. A research question leads to a set of hypotheses that need not be mutually exclusive but should all be testable. Each hypothesis should generate one or more specific predictions.

Statistical analysis is usually necessary to answer questions with behavioural data. Analysis should be planned and registered before collecting data. Once collected, a dataset should be formatted and permanently archived prior to analysis. Data is checked and visualised with descriptive statistics and graphs. Models representing hypotheses about the true effects present in the population from which the dataset is a sample are built and tested with inferential statistics. Many different hypotheses can be captured using a linear modelling framework in which an outcome variable is predicted with a combination of predictor variables and interactions. Sources of non-independence in datasets can be addressed with mixed models. The robustness of findings can be examined by comparing the results obtained when analysis is done in different ways using model selection and multiverse approaches. Confirmatory analysis designed to test preregistered hypotheses should be clearly differentiated from exploratory analysis that generates new hypotheses.

Measuring behaviour means assigning numbers to observations of behaviour according to specified rules. Converting a stream of behaviour into behavioural metrics involves choosing and defining specific categories of behaviour that can be measured. Behavioural categories can be described in terms of their physical structure or their consequences. An ethogram is a catalogue of the species-typical behavioural categories displayed by a species in a specified environment. Descriptions of behavioural categories should be unambiguous and written down before data collection starts. Behavioural categories can be designated as either events (short duration) or states (longer duration). Behavioural categories are used to generate metrics such as latencies, frequencies, durations and intensities. Two or more metrics can be combined to form a composite metric. Metrics can be at different levels of measurement, ranging from nominal (weakest) to ratio (strongest).

Poor-quality measurements are likely to yield meaningless or unrepeatable findings. High-quality measurements are characterised by validity and reliability. Validity relates to whether the right quantity is measured and is assessed by comparing a metric with a gold-standard metric. Reliability relates to whether measurements are repeatable and is assessed by comparing repeated measurements. The accuracy and precision with which measurements are made affect both validity and reliability. A major source of unreliability in behavioural data comes from the involvement of human observers in the measurement process. Where trade-offs are necessary, it is better to measure the right quantity somewhat unreliably than to measure the wrong quantity very reliably. Floor and ceiling effects can make measurements useless for answering a question, even if they are valid and reliable. Outlying data points should only be removed if they can be proved to be biologically impossible or to result from errors.

Interpreting results correctly and communicating them honestly are vital parts of what scientists do. Incorrect interpretation of data often results from avoidable statistical mistakes. Common pitfalls arise from abuse of significance testing, misunderstanding of correlations and overgeneralisation of findings. Publishing peer-reviewed papers in scientific journals is the primary means by which researchers communicate their findings to other scientists. A scientific paper has an established basic format comprising title, abstract, introduction, methods, results and discussion. Open Science practices are an important part of the modern publication process. Non-technical (lay) summaries and press releases are tools for communicating behavioural research to journalists and the public. All science involves potential conflicts of interest, and their influence on scientific communication is an unresolved cause for concern. Several organisations oversee the integrity of science, but ultimately it is the personal responsibility of each individual researcher to behave with openness and integrity.

Public trust in science depends on scientists behaving legally and ethically. Ethical science is also often better science. To be ethical, research must be of sufficient quality to further scientific understanding and its potential benefits should outweigh the risks of harm to subjects or other stakeholders. All research must also be lawful. Conducting a harm–benefit analysis is central to ensuring that ethical standards are maintained in research and is required for the majority of behavioural studies. Formal ethical approval must be obtained before starting to collect data. Research on animals should minimise animal suffering by following the 3Rs principles of replacement, reduction and refinement. Humane end points should be used to limit unnecessary suffering. Research on humans should respect the autonomy and rights of participants and will generally require informed consent, the right to withdraw and debriefing. Deception is potentially harmful and should only be used following careful consideration.

High-quality behavioural data can be recorded using cheap and simple technologies such as checks sheets and sound recorders. Advances in technologies for data recording have made big data available to behavioural scientists, which in turn has stimulated the development of AI technologies for automated data processing. A data pipeline describes the workflow of data recording, processing and analysis, including details of the technologies used in each step. The choice of technology for capturing behavioural data will depend on the research question and the resources available, the quantity of data required, where the data is to be collected, the amount of interaction with subjects and the likely impact of the technology on the subjects and their environment. Data that are initially recorded in a relatively rich form will require subsequent processing to code behavioural metrics. Coding of data can be either manual or automated using rules-based approaches and machine learning.

Behavioural studies aim to discover scientific truths. True facts should be replicable, meaning that the same conclusions are reached if the same data are analysed, if the same methods are applied to collect a new dataset and if different methodological approaches are used to address the same general hypothesis. The replication crisis refers to a widespread failure to replicate published findings in the biological and social sciences. The causes of the replication crisis include the presence of uncontrolled moderators of behaviour, low statistical power and dubious research practices. Various sources of information can help to distinguish good research from bad. An evidence pyramid ranks different study types according to the quality of evidence produced. The Open Science movement encourages replication, preregistration and transparency over materials, methods and data, all of which should improve the quality of science and the likelihood that findings will be replicated.

Behaviour is the actions and reactions of an organism or group of organisms. Living organisms, robots and virtual agents all exhibit measurable forms of behaviour. Measuring behaviour involves assigning numbers to direct observations of behaviour using specified rules. Direct observation means collecting data that relates directly to the performance of the behaviour pattern in question. Measuring behaviour accurately and reliably is important because behaviour is central to answering many questions in the biological and social sciences. Measuring behaviour is challenging because behaviour has a temporal component, does not always occur in discrete bouts, is generally complicated, can be influenced by stimuli undetectable to humans and varies both within and between individuals. Studying behaviour can be broken down into a series of steps that starts with asking a question and ends with communicating findings.

Social behaviour can be measured at different levels, from the behaviour of individuals to the behaviour of very large groups. The group is the basic unit of social organisation and must be clearly defined. It will often be important to measure group size. Crowding describes the average group size experienced by an individual. Individual identification is essential in many studies and can be accomplished either by artificially marking or tagging individuals, or by using natural variation. Marking and tagging have ethical and scientific implications. Social network analysis is the set of methods for describing and analysing how individuals interact within a group. Social network analysis yields metrics that describe properties of social interactions at both the individual and group levels. Dominance hierarchies rank the individuals within a group relative to one another and can be characterised in terms of their linearity, steepness and temporal stability.

Study design is fundamental to good science. A poorly designed study will waste time and resources and could produce misleading or uninterpretable results. A good study design aims to minimise random variation and eliminate confounding variables. Correlational studies make use of natural variation in the variables of interest, whereas experimental studies manipulate variables to understand their causal effects on behaviour. There are advantages and disadvantages to both types of study, but experiments uniquely allow inferences about causation. A good experimental design requires subjects to be randomly allocated to experimental groups. Randomisation ensures the generalisability of results and eliminates confounds in experimental studies. Measurements should ideally be made blind to group membership. Blinding minimises biases caused by the conscious or unconscious expectations of the experimenter or subjects. Careful consideration should be given to when behaviour is measured, as time can affect behaviour. Power calculations can be used to determine the appropriate sample size.

Behaviour can be recorded in either the laboratory or the field. In either setting, it can be recorded using standardised behavioural tests that elicit specific behaviour, or by observing freely-behaving subjects. Observation requires decisions about which subjects to observe (sampling rules) and how to record their behaviour (recording rules). There are four sampling rules: ad libitum sampling, focal sampling, scan sampling and behaviour sampling. There are two basic types of recording rule: continuous recording and time sampling; the latter can be further divided into instantaneous sampling and one–zero sampling. Continuous recording is more demanding for the observer but is the only recording method that produces true frequencies and durations. Estimates of frequencies and durations derived from time sampling will be more accurate if the sample interval is short relative to the mean duration of the behaviour. One–zero sampling is likely to yield biased estimates of frequency and duration.