Ethology is the biological study of behaviour: what an animal does, how its actions are produced, how they develop, and how they relate to evolutionary history and life in a particular environment. The gold standard is an explanation that connects a precisely described action with evidence capable of distinguishing competing causes. A memorable animal story becomes science when another observer can check the behaviour and the reasoning behind the claim.
This guide helps you move from watching an action to understanding its organisation. You will learn to separate stimulus, sensory access, internal state and response; distinguish several kinds of learning; analyse sequences and observation records; and recognise what an experiment actually establishes. The numerical datasets and teaching scenarios below are original, invented examples. They illustrate reasoning, not measured characteristics of a species or instructions for conducting experiments on animals.
Begin with the Gold Standard subject hub for the wider collection. The purpose follows The Gold Standard of Education: an explanation should remain useful when the organism, measurement or circumstances change. For behaviour’s ecological costs, benefits and evolutionary trade-offs, continue into the complementary Behavioural Ecology guide.
THE GOLD STANDARD · ETHOLOGY
Follow behaviour from observation to explanation
Choose a route through behavioural mechanisms, learning and the measurements needed to distinguish explanations. All teaching chapters remain open below.
ROUTE 1 · CHAPTERS 1–4
Define behaviour and its causes
Connect observable actions, explanatory questions, sensory access and state.
ROUTE 2 · CHAPTERS 5–8
Follow development and learning
Separate acquisition, performance, habituation and contingency.
ROUTE 3 · CHAPTERS 9–12
Read interactions and observations
Examine social learning, communication, ethograms and denominators.
ROUTE 4 · CHAPTERS 13–16
Analyse sequences and evidence
Work through transitions, agreement, comparisons and automated tracking.
ROUTE 5 · CHAPTERS 17–20
Diagnose, transfer and continue
Repair inferences, solve a changed-context case and plan the next learning step.
Full chapter index · Try the transfer task · Gold Standard subject hub
Chapter index
Define behaviour and its causes · Chapters 1–4
Follow development and learning · Chapters 5–8
Read interactions and observations · Chapters 9–12
Analyse sequences and evidence · Chapters 13–16
Diagnose, transfer and continue · Chapters 17–20
CHAPTER 1 OF 20 · Define behaviour and its causes
1. Start with an action that another observer could recognise
Imagine a recording of an animal turning its head after a sound. “It became curious” proposes an interpretation. “Its head rotated towards the speaker within two seconds” describes something that can be checked, provided the recording shows the relevant movement and the sound onset. Both sentences may be interesting, but they do different jobs.
A useful behavioural description identifies the individual, action, setting and time. It also defines the boundary of the event. Does a small head movement count? Must the animal face the speaker? What happens if its head was already pointing that way? Without those decisions, two observers can produce different results while each thinks the task is obvious.
Begin a notebook with separate columns for observation and interpretation. In the observation column, record the action and what immediately preceded and followed it. In the interpretation column, write possible explanations and the evidence each would require. A third column for missing information prevents an attractive explanation from silently filling a gap in the recording.
Precision need not make the animal seem mechanical or uninteresting. It protects the question. If a behaviour might express attention, memory or a social relationship, careful description allows those possibilities to be investigated instead of accepted or rejected by intuition.
The existing Primate Behaviour Vocabulary guide provides a useful language route through actions, postures and interactions. Use its terms to sharpen a description, while checking that each term fits the species and evidence in front of you.
CHAPTER 2 OF 20 · Define behaviour and its causes
2. Ask which kind of explanation you need
Tinbergen’s 1963 paper on the aims and methods of ethology distinguishes four connected questions: causation, development, survival value and evolution. In contemporary teaching these are often expressed as mechanism, ontogeny, adaptive function and evolutionary history. They organise different kinds of evidence rather than four competing answers from which only one can be correct.
Consider the invented head-turning observation. A mechanism question asks which sensory information and bodily processes produced the turn. A development question asks how that response changes with age and experience. A function question asks whether the response has consequences relevant to survival or reproduction. An evolutionary question asks how related forms of behaviour arose and changed across lineages.
Now test an explanation: “The animal turns because locating a sound can help it find food.” This proposes a possible benefit. It does not identify the sensory pathway that generated this particular movement. Conversely, describing a neural circuit does not demonstrate that the response improves reproductive success in the wild.
Write the intended question above any diagram or calculation. Then ask whether the evidence belongs to that question. This simple step catches many errors that otherwise survive several pages of technical language.
The questions also guide collaboration. A field observer, developmental researcher and neuroscientist may examine the same behaviour using different measurements. Their findings become more useful when the connection between levels is explicit. Agreement on a behaviour’s name is not sufficient; they must also establish that their tasks measure comparable actions under comparable conditions.
CHAPTER 3 OF 20 · Define behaviour and its causes
3. Follow the path from environment to movement
A minimal mechanism has several links: an environmental event reaches a sensory system; the organism processes that information in its current state; activity in motor systems changes the body; the movement changes what the organism encounters next. The final step matters because behaviour can alter its own input.
For example, turning towards an object may bring it into a different part of the visual field. Moving closer may increase the strength of a sound or chemical signal. A continuing action can therefore depend on feedback rather than on a single initial command.
Use a diagram to expose what is known at each link. In an invented video study, the visible record might establish that an object moved and the animal subsequently turned. It might not establish whether the animal detected the object’s movement visually, heard its motor or responded to another animal nearby. The explanation has a missing sensory link.
The repair is to identify observations or controlled comparisons that would distinguish those routes. It is not enough to draw an arrow labelled “perception” between the object and the action. That arrow represents a research question until the relevant evidence exists.
The Neuroscience root develops the connection between sensory processing, circuits and movement. Use it when the question concerns nervous-system mechanisms. Return to the whole animal afterwards: a circuit operates inside a body, in a setting, with a history of experience. A mechanism that explains one laboratory response may need additional information to explain the same animal’s behaviour elsewhere.
CHAPTER 4 OF 20 · Define behaviour and its causes
4. A stimulus is only meaningful through sensory access and state
“The same stimulus was presented” can conceal several differences. The equipment may have produced the same output while the animal occupied a different position, faced another direction or encountered a changed background. What the observer calls the stimulus and what reaches the animal’s receptors are related but not identical quantities.
Suppose an invented camera record contains two sessions with the same moving target. In one, the animal faces an unobstructed area; in the other, a screen blocks its view for half the session. A lower turning count in the second session cannot immediately be attributed to weaker motivation. Sensory opportunity changed first.
Internal conditions also matter. Age, reproductive condition, recent activity and physiological state can affect behaviour. These are candidate explanations, not permission to diagnose an invisible state from a single movement. Record what is measured separately from what is inferred.
The Physiology guide explains how bodily systems coordinate function. It helps distinguish a proposed internal mechanism from an informal label such as “more energetic”. The Chronobiology root adds another relevant condition: observations at different biological times may not be directly comparable.
A practical question for every comparison is: could both animals detect the event, perform the action and receive the same opportunity? If one of these conditions is uncertain, the observed difference remains real, but its interpretation becomes narrower. Repair the uncertainty before turning a performance difference into a claim about intelligence, preference or temperament.
CHAPTER 5 OF 20 · Follow development and learning
5. Development connects inherited organisation with experience
The question “Is this behaviour innate or learned?” is often too coarse to explain how it develops. An animal inherits biological resources that develop in an environment. Experience can alter some aspects of performance while others remain relatively stable. Different components of the same behaviour may depend on different developmental processes.
Consider an invented comparison of young and older animals solving a visual discrimination task. Older animals perform better. They may have had more relevant experience, more mature sensory capacities, greater motor control or a different willingness to engage. Age alone does not separate these possibilities.
A developmental account follows changes through time and records the conditions under which they occur. Repeated observations of the same individuals can reveal trajectories, although the observations or repeated tasks may themselves provide experience. Comparing separate age groups avoids that particular repetition but introduces differences in individual history. The design determines which uncertainty remains.
Do not interpret an early appearance as proof that experience is irrelevant. Nor does a measurable effect of experience show that biology places no constraints on learning. Both claims go beyond the comparison.
For a learner, the useful repair is to replace the binary label with a component question. Ask whether experience changes recognition of a cue, the timing of an action, the precision of movement or selection between alternatives. Each component suggests a different measurement. This approach turns development into a sequence of investigable changes instead of a contest between two broad words.
CHAPTER 6 OF 20 · Follow development and learning
6. Birdsong shows why acquisition and performance must be separated
A published example makes the developmental problem concrete. In a 2021 zebra finch study, Liu and colleagues examined juveniles’ attentive listening and approach towards a singing tutor. The study connected earlier exposure, later engagement with the tutor and the precision of song imitation. Its findings support an active role for continued auditory engagement during song development; they do not establish one universal timetable for every songbird.
The general reasoning task is to separate access to a model, formation of a memory, production of an action and later refinement. Failure at the final performance could arise at more than one of these stages. A juvenile that does not reproduce a particular sound has not thereby demonstrated that it never detected or retained information about it.
Imagine a fictional learning record with good recognition of a model but imprecise production. Increasing exposure alone might not address the limiting step. Conversely, fluent production of a familiar sequence does not establish flexible recognition of altered examples. The next measurement should target the missing connection.
This distinction is useful when reading any animal-learning paper. Identify what the task makes observable. A choice test, movement trace and vocal recording provide different evidence. Ask whether the authors measured acquisition directly or inferred it from later performance, and whether an alternative sensory or motor explanation was considered.
When connecting such research with human learning, preserve the boundary between analogy and evidence. Birdsong can clarify a question about learning stages. It does not, by itself, validate a classroom method or prescribe a child’s developmental schedule.
CHAPTER 7 OF 20 · Follow development and learning
7. Reduced responding is not automatically habituation
Habituation describes a decrease in responding with repeated stimulation that is not simply sensory adaptation or motor fatigue. The Rankin and colleagues consensus paper discusses behavioural characteristics used to identify it, including stimulus specificity and recovery patterns. A declining response curve alone is insufficient to establish the process.
Imagine a fictional archive in which an animal turns after eight of ten early events but after only three of ten later events. The arithmetic shows a reduction from 80% to 30%, or 50 percentage points. It does not identify the cause.
The camera may have lost the relevant angle. Background noise may have changed. The animal may no longer be able to perform the scored movement in the same way. A learning explanation is one possibility among several.
Before interpreting the record, check whether event delivery, sensory access, scoring and opportunity remained comparable. Then inspect the study’s evidence for distinguishing learning from other causes of response loss. Recovery or stimulus-specific responding can contribute evidence, but no single informal observation settles every alternative.
For school practice, analyse an existing dataset or permitted recording rather than repeatedly provoking wildlife. The important skill is to explain why the descending line supports a descriptive claim more directly than it supports a particular biological mechanism.
CHAPTER 8 OF 20 · Follow development and learning
8. Learning about events differs from learning about consequences
In associative learning, an organism can acquire relationships among events or between its actions and their consequences. These are useful distinctions because the relevant comparison changes. A cue that predicts an outcome is different from an outcome that depends on performing an action.
Rescorla’s 1968 experiments demonstrated why event contingency matters in conditioning: the probability of an outcome outside the cue period also affects what can be learned about the cue. Counting cue–outcome pairings alone can miss the informative relationship.
Consider this separate, invented calculation. In a dataset divided into comparable observation windows, an outcome occurs in 16 of 20 cue-present windows and 4 of 20 cue-absent windows. The estimated proportions are 0.80 and 0.20; their difference is 0.60. In another dataset the outcome occurs in 16 of 20 windows in both conditions. The pairing count during the cue is unchanged, but the difference is now zero.
That calculation describes the recorded contingency. It is not a complete learning model, and it does not prove that the animal detected the cue or used the information. Timing, previous experience and the task still matter.
For action–consequence questions, ask a different counterfactual: would the outcome have occurred if the action had not happened? A rewarding event that merely follows a movement may have been independently scheduled. A credible interpretation needs the contingency and a relevant comparison, not only chronological order.
CHAPTER 9 OF 20 · Read interactions and observations
9. Social learning needs evidence beyond animals doing the same thing
Animals can resemble one another because they share a setting, encounter the same opportunities or independently discover similar solutions. Similarity alone does not establish that one learned from another. The question is whether information obtained through another individual influenced learning.
In Bridges and colleagues’ 2023 bumblebee study, demonstrators used alternative solutions to a puzzle box, and those variants spread through groups. The two-action design and groups without demonstrators helped distinguish social transmission from individual discovery. Some bees opened boxes without demonstrators, which makes the control comparison especially informative.
Do not collapse social learning, imitation and teaching into one claim. An observer might become interested in a location or object without copying the demonstrator’s detailed movement. Calling an interaction “teaching” introduces further requirements about what the demonstrator does and how that affects learning.
For a fictional example, suppose observers later choose the same coloured object as a demonstrator. That result raises a question about transmitted information, but it does not reveal whether they copied a movement, learned a location or developed an attraction to the colour. Changing those features in a suitably designed comparison could help separate the explanations.
The transferable skill is to identify exactly what was shared: attention, a destination, an object, an outcome or an action pattern. A compelling video may show the final similarity while omitting the comparison needed to explain how it arose.
CHAPTER 10 OF 20 · Read interactions and observations
10. Communication requires a sender, receiver and context
An animal’s movement or sound can be observed without its meaning being established. A communication claim needs attention to the producer, the receiver, the form of the event and the surrounding interaction. A response by the receiver can be informative, but it is not a literal translation into human language.
Consider a fictional recording in which one animal calls and another moves away. Several events may coincide: the caller moves closer, a third individual appears or an environmental disturbance occurs. The sequence suggests a question about the call; it does not isolate the call as the cause of withdrawal.
Begin with a timeline that preserves all visible events. Then distinguish the acoustic event from other changes. Ask whether the same receiver reacts similarly across contexts and whether different calls are associated with different responses. Meaning cannot be assigned by selecting the most dramatic occurrence and ignoring the rest.
Also distinguish a signal from any feature that another animal can use as information. An incidental sound produced during movement may reveal location without having evolved as a communication signal. Establishing the evolutionary role requires evidence beyond the observer’s ability to detect the feature.
For ecological questions about signal costs, conflicts of interest and consequences, the existing How Science Works: Behavioural Ecology article provides the connected route. Here the immediate discipline is to keep the recorded event, the receiver’s measured response and the proposed function separate until the evidence joins them.
CHAPTER 11 OF 20 · Read interactions and observations
11. Build an ethogram around the question
An ethogram is a defined catalogue of behaviours used in an observation. It is a measurement instrument, not simply a list of interesting verbs. Its categories must fit the question and the visibility of the record.
For an invented video exercise, suppose the question concerns transitions between stationary behaviour and movement. “Stationary”, “walking” and “out of view” could be useful primary categories. “Happy”, “bored” and “determined” would import interpretations that the video may not support.
Decide whether categories are mutually exclusive. Walking and calling can occur together; a single activity column cannot represent both without a stated priority rule. One solution is separate movement and vocalisation channels. Another is a combined coding system designed for a specific question. Neither choice is automatically best, but the choice affects what can be calculated afterwards.
Define onset, offset and uncertain cases. If a movement stops for half a second, does that end a bout? How much obscuration makes the animal unscorable? Pilot the definitions on material separate from the final comparison, then revise ambiguous boundaries before scoring the main record.
Altmann’s foundational sampling paper explains why the observation method influences the evidence obtained. Following one focal individual and scanning a group are different sampling choices; recording continuously and scoring at selected instants answer different questions.
The practical check is to hand the definition sheet and an unfamiliar clip to another observer. If they cannot apply the categories without asking what you meant, the next improvement belongs in the definitions rather than in a more elaborate statistical test.
CHAPTER 12 OF 20 · Read interactions and observations
12. Choose the denominator before comparing response rates
Suppose two invented sessions contain 12 and 18 head turns. Which session shows more responding? The second has the larger count. That answer changes if the sessions supplied different observation time or different numbers of detectable events.
| Quantity | Session A | Session B |
|---|---|---|
| Scheduled recording | 10 minutes | 10 minutes |
| Scorable animal time | 8 minutes | 5 minutes |
| Eligible cue events | 20 | 40 |
| Scored turns after eligible cues | 12 | 18 |
Turns per scorable minute are 12 ÷ 8 = 1.5 for A and 18 ÷ 5 = 3.6 for B. Response proportions per eligible cue are 12 ÷ 20 = 0.60 for A and 18 ÷ 40 = 0.45 for B. B has more turns per observed minute but a lower proportion of cue events followed by a turn.
Neither calculation invalidates the other. They answer different questions. Event rate includes how often opportunities occur. Conditional response proportion asks what happened when a defined opportunity was available.
The validity of the second calculation depends on “eligible”. In this example, eligibility requires a visible animal in a position where the cue could be detected, a complete response window and no overlap with another cue. Counting invisible or overlapping opportunities would alter the denominator without adding comparable evidence.
Report missing observation separately. Eight scorable minutes out of ten does not establish what happened in the remaining two. If disappearance is related to the behaviour, even a correctly calculated visible-time rate can be an unrepresentative estimate of the entire session. The arithmetic is transparent; the sampling limitation must be equally visible.
CHAPTER 13 OF 20 · Analyse sequences and evidence
13. Behavioural order contains information that totals lose
Two recordings can contain the same number of approaches, pauses and withdrawals while organising them differently. A total count discards the order. Sequence analysis asks what tends to follow what, within a defined timescale and coding rule.
Imagine 100 scored transitions pooled from several short fictional clips. Each transition is between successive coded states at a fixed sampling interval; no transition is counted across a break between clips. The rows below identify the current state and the columns identify the next one. Repeated states are retained. Because the clips have separate starts and ends, the totals entering a state need not equal the totals leaving it.
| Current state | Next: stationary | Next: walking | Row total |
|---|---|---|---|
| Stationary | 36 | 24 | 60 |
| Walking | 10 | 30 | 40 |
The estimated probability of walking next, given stationary now, is 24 ÷ 60 = 0.40. Given walking now, it is 30 ÷ 40 = 0.75. The record therefore contains persistence in walking at this chosen interval. Dividing 24 by 100 would answer a different question: the fraction of all recorded transitions that were stationary-to-walking.
If these probabilities were stable and the next state depended only on the current one, a two-state Markov model would be a possible simplification. At equilibrium, the fraction walking, w, would satisfy w = 0.40(1 − w) + 0.75w. Therefore 0.65w = 0.40, giving w ≈ 0.615, or 61.5%.
That is a model prediction under assumptions, not the observed time budget. Real transitions may depend on earlier events, time already spent in a state or changes in the environment. A different sampling interval can also change the table. Check those possibilities before treating the two-state model as an account of the mechanism.
CHAPTER 14 OF 20 · Analyse sequences and evidence
14. Observer agreement is a property to measure
Suppose two observers independently score 100 predetermined video instants as walking or not walking. They both score walking at 30 instants and both score not walking at 50. At 10 instants only the first scores walking; at another 10 only the second does.
Their observed agreement is (30 + 50) ÷ 100 = 0.80, or 80%. Both observers classify 40% of the instants as walking and 60% as not walking. Under a comparison based on independent assignments with those marginal proportions, expected agreement is (0.40 × 0.40) + (0.60 × 0.60) = 0.52.
Cohen’s kappa expresses agreement beyond that baseline relative to the available improvement: (0.80 − 0.52) ÷ (1 − 0.52) ≈ 0.583. It is a useful descriptive calculation here, not a universal certificate of acceptable coding. The category frequencies, disagreement pattern and intended use still matter.
Inspect the actual disagreements. Perhaps one observer counts a foot repositioning as walking and the other requires displacement of the body. A revised boundary, illustrated with examples and counterexamples, could repair the measurement more directly than simply averaging the two scores.
Agreement also differs from accuracy. Two observers can share the same mistaken rule. An automated classifier trained on their labels can reproduce that mistake consistently. Where possible, compare the coding against a defensible reference and examine the difficult cases separately.
Finally, the 100 instants may come from a small number of animals or neighbouring moments in the same clip. They describe coding agreement on that material. They do not automatically provide 100 independent biological replicates or justify a narrow confidence interval for a species-wide claim.
CHAPTER 15 OF 20 · Analyse sequences and evidence
15. A controlled comparison should isolate the proposed cause
An intervention can strengthen causal reasoning when the comparison separates the proposed cause from other systematic differences. The useful question is not merely whether two groups differ. It is whether the design makes the alternative explanations less plausible.
Imagine a fictional study comparing responses to two visual patterns. All observations of pattern A happen in the morning, and all observations of pattern B happen in the afternoon. A difference could reflect pattern, time, order, recent experience or some combination. Increasing the number of repeated trials without changing that arrangement does not remove the confounding.
Random assignment, balanced orders, suitable controls and blinded scoring can address different problems. Random assignment concerns which treatment an experimental unit receives. Blinding concerns what the observer knows while making a judgement. Neither replaces a definition of the experimental unit.
If a treatment is applied to a whole enclosure, animals within that enclosure share the treatment environment. Treating every frame from every animal as an independent replicate exaggerates the amount of independent evidence. Analysis should respect the level of assignment and the dependence within individuals and groups.
Evolutionary comparisons introduce another dependence: related species share ancestry. Similar behaviour in two closely related species does not provide the same independent evidence as two unrelated origins under comparable conditions. A persuasive evolutionary explanation needs an appropriate comparative design, not a list of similarities selected after the fact.
For a reader, the repair is to sketch what was varied, what was measured and what remained linked to the variation. If two candidate causes always move together, the study may not be able to separate them, however precise the measurements become.
CHAPTER 16 OF 20 · Analyse sequences and evidence
16. Automated tracking measures movement before it explains behaviour
Tools such as DeepLabCut estimate user-defined body-part positions from images. This can make detailed movement records feasible. The output is still an estimate of locations; identifying a biological action and interpreting its function require further definitions and validation.
Consider a fictional tracker that labels an animal’s head and tail. When two individuals cross, the identities briefly swap. The resulting path may show an implausibly rapid movement. A summary that converts every position difference into speed will turn the tracking error into a behavioural event unless the pipeline catches it.
Validate difficult conditions deliberately: overlap, changes of posture, partial obscuration, lighting differences and unfamiliar backgrounds. An average error from easy frames can conceal failure precisely where the behaviour of interest occurs. Keep a held-out set representative of the material on which the result will be used.
Also avoid testing only on frames immediately adjacent to training frames. Such frames may be almost identical. A model can appear to generalise while depending on the same individual, camera arrangement or background. The validation split should reflect the intended claim: new moments, new individuals or a new setting require different checks.
Preserve raw recordings, annotation rules, model version and exclusions so a result can be reconstructed. If a low-confidence point is removed, record why and how missing values affect later calculations. Interpolation can be helpful for some tasks, but it can also invent a smooth path across an unobserved action.
Automation is most useful when it makes a defined measurement more consistent and inspectable. A label such as “aggression detected” needs more justification than a set of tracked coordinates.
CHAPTER 17 OF 20 · Diagnose, transfer and continue
17. Diagnose the inference that failed
When a behavioural explanation goes wrong, identify the particular move that exceeded the evidence. Different errors need different repairs. Collecting more of the same observations can make a mistaken inference look more confident without making it more defensible.
A motive has replaced a description
“The animal is jealous” may be an engaging story, but the immediate repair is a record of who approached, who moved away, what contact occurred and which context preceded the interaction. An interpretation of affect requires additional evidence and a clear account of alternative explanations.
A low score has become a low-ability label
Check sensory access, motor demands, opportunity and willingness to engage. A task can fail to measure the capability it was intended to measure. The repair is a more appropriate measurement, not a flattering reinterpretation of every poor result.
An absent observation has become an absent behaviour
Check the field of view, recording period, detection threshold and missing intervals. “Not recorded during this session” is often the warranted statement. It does not imply that every conceivable unobserved event happened; it defines the evidence boundary.
A group mean has become a typical individual
Inspect the distribution and repeated measurements. A mean difference might reflect a general shift, a few extreme individuals or unequal contributions from prolific participants. Those possibilities imply different biological accounts.
A functional story has become a demonstrated adaptation
An action may appear useful without the proposed benefit having been measured or its evolutionary history established. Preserve the hypothesis, then identify the fitness or comparative evidence needed to test it. A plausible story is a starting point for inquiry, not the final result.
CHAPTER 18 OF 20 · Diagnose, transfer and continue
18. Transfer task: the impressive result changes when opportunities change
An invented observational dataset compares two groups viewing a naturally occurring visual event on archived recordings. Group A contains ten animals, each with ten eligible event windows. Group B contains ten animals, each with five eligible windows. In A, 60 windows contain the defined head turn. In B, 35 contain it. Assume each window has at most one scored response and all listed windows meet the same visibility rule.
A report says: “Group A produced more responses, so it was more responsive.” Calculate the response proportions and explain the problem with that conclusion.
Now add a complication. In A, five animals respond in all ten windows and five respond in two. In B, every animal responds in either three or four of its five windows: five animals respond in three and five in four. What has the group average hidden?
Finally, the camera loses sight of several A animals during another 40 event windows, which were excluded as ineligible. Explain why the visible-window result cannot settle a claim about all events encountered by the animals. Do not invent responses for the missing windows.
Answer and reasoning
A has 10 × 10 = 100 eligible windows, giving 60 ÷ 100 = 60%. B has 10 × 5 = 50, giving 35 ÷ 50 = 70%. A produced more recorded turns, but B had the higher response proportion per eligible event. “More responsive” needs an explicit measurement; raw totals are inadequate when opportunities differ.
For A, the individual proportions are five values of 100% and five of 20%. Their equally weighted mean is 60%, but no individual has that value. For B, the proportions are five values of 60% and five of 80%, averaging 70%. A has much greater between-individual variation in this constructed record. A single mean conceals that difference.
The windows are repeated observations within animals. They should not be treated as 150 independent animals. Nor does this observational comparison, without further design information, establish the cause of the difference between groups.
The missing 40 windows create a separate uncertainty. If all had contained no response, A’s proportion across the 140 recorded-or-missing event windows would be 60 ÷ 140 ≈ 42.9%. If all had contained a response, it would be 100 ÷ 140 ≈ 71.4%. These are extreme bounds, not estimates. They show why the incomplete record cannot determine the full-event proportion or its ranking against B’s 70%.
The useful conclusion is precise: B has the higher observed proportion under the shared eligibility rule, A is more heterogeneous in the supplied individual records, and the excluded windows prevent a confident claim about A’s full-event response proportion.
Is ethology the same as behavioural ecology?
They overlap. Ethology is the broad biological study of behaviour, including mechanisms, development, function and history. Behavioural ecology concentrates on behaviour in relation to ecological conditions and evolutionary consequences. A question can use both perspectives, provided the evidence for each is clear.
Does observing in the wild automatically make a study better?
No setting answers every question. Natural observation can preserve important context but leave causes difficult to separate. A controlled setting can isolate a relationship while changing the conditions under which it operates. Judge the fit between question, setting and inference, then ask what would support transfer between settings.
Can animals have emotions if we avoid emotional labels in scoring?
Avoiding an unsupported label is a measurement decision, not a denial of animal experience. Research on affect needs converging evidence and species-appropriate interpretation. A coding scheme should distinguish what is directly measured from the broader state proposed to explain it.
Can I practise without handling animals?
Yes. Use permitted archival recordings, published datasets or the invented examples here. Define a small ethogram, score an unfamiliar clip, compare independent coding and explain the limits. These activities practise the central reasoning without changing an animal’s circumstances.
Does statistical significance establish a biological mechanism?
No. A statistical result concerns a model and data under assumptions. A mechanism claim also depends on the design, measurement and competing explanations. Report the effect’s size and uncertainty, the unit of replication and the biological interpretation separately.
Is one clever animal enough to establish a species-wide capacity?
One individual may demonstrate that a performance occurred under particular conditions. Estimating how common it is, explaining how it developed or attributing it to an entire species requires additional evidence. Preserve the interesting observation while keeping its scope accurate.
CHAPTER 20 OF 20 · Diagnose, transfer and continue
20. Build a learning route that ends with a defensible explanation
Start with a short permitted recording. Write three descriptions that another observer could check. Remove motive words from the observation column, then keep the corresponding hypotheses in a separate interpretation column. The aim is to preserve curiosity while making the evidence traceable.
Next, choose one question and define no more categories than it needs. Pilot them, clarify the difficult boundaries and score fresh material. Calculate one quantity with its denominator and units. Explain why another plausible denominator would answer a different question.
Then examine a sequence or a comparison. State the experimental or observational unit, identify repeated measurements and draw the alternative causal explanations. Finish with a changed-context test: what would need checking if the lighting, species, developmental stage or observation window changed?
In practical work, behavioural evidence can inform welfare assessment, conservation monitoring and the design of animal environments. These applications require relevant expertise and oversight. A classroom coding exercise is not a validated welfare assessment or permission to manipulate animals. The Royal Society’s research ethics guidance describes expectations for animal welfare, review and transparent reporting in research; consult the applicable institutional and local requirements for any real project.
Follow the question into the connected library
- For ecological choices, costs and consequences, read The Gold Standard of Behavioural Ecology and its How Science Works specialist article.
- For sensory processing and action, continue to Neuroscience; for bodily coordination, choose Physiology.
- For biological timing, follow Chronobiology and How Science Works: Chronobiology.
- For precise language about social actions, use the Primate Behaviour Vocabulary guide.
- To explore another mechanism, return to How X Works, or choose another broad subject from The Gold Standard.
The linked research provides distinct source routes: Tinbergen for the explanatory questions, Altmann for observation methods, Rankin and colleagues for habituation criteria, Rescorla for contingency, Liu and colleagues for song development, and Bridges and colleagues for social learning. Read the methods as well as the conclusions. The original calculations in this guide show how to interrogate a record; they do not replace the evidence of those studies.
A successful final explanation names the behaviour, identifies the question, shows the measurement, compares plausible causes and states what remains unresolved. That is a capability you can carry from one animal story to the next.