Reading for lecture 2 - Recording Methods in Behavioural Observation

Publication & Context

• Third Edition of Measuring Behaviour – An Introductory Guide, Cambridge University Press (CUP).
• Authors: Paul Martin & Patrick Bateson.
• Key bibliographic facts: Hardback ISBN $978\text{–}0\text{–}521\text{–}82868\text{–}0$; Paperback ISBN $978\text{–}0\text{–}521\text{–}53563\text{–}2$.
• First published $2007$; $8^{\text{th}}$ printing $2013$.
• CUP locations: Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Mexico City.
• CUP disclaims responsibility for accuracy/persistence of external URLs & time-sensitive factual data.

Chapter 5 Focus: Recording Methods

• Two levels of methodological choice:
– Sampling Rules → decide which subject(s) to watch and when.
– Recording Rules → decide how the behaviour is actually registered.
• Never confuse a sampling rule with a recording rule (e.g. “focal sampling” ≠ continuous recording).

Sampling Rules ( “Which/When” )

1 Ad libitum Sampling

• Observer writes down whatever seems relevant whenever visible.
• Strengths: good for preliminary work; captures very rare, salient events.
• Limitations/Biases:
– Skews toward conspicuous individuals/behaviours; misses brief responses; under-represents some ages (Hernández-Lloreda $2006$).

2 Focal Sampling ( “Focal Animal/Dyad/Litter” )

• Observe one predefined unit for a fixed duration; record all behaviour categories.
• Order of focal subjects should be randomized/systematic across sessions.
• Social context: must also log interactants, initiators, recipients.
• Interruptions: if focal subject goes out of view, mark “time out”; final metrics must be corrected for visible time.
• Special field dilemma: whether to pursue vanished subjects & related explicit rules.

3 Scan Sampling

• Rapidly "census" whole group at regular intervals; register each individual’s state at that instant.
• Usually limited to simple categories (e.g. asleep? feeding? nearest neighbour?).
• Duration per individual should be negligible/constant.
• Scan length may span $\text{few s} \Rightarrow \text{several min}$ depending on group size & variables.
• Biases: conspicuous individuals/acts over-represented.
• Practical merits: ensures balanced data across subjects, time & season (e.g. de Ruiter $1986$ on capuchins).
• Can combine with focal work (e.g. focal detail + group scan every $10$–$20$ min).
• Statistical independence: scans must be sufficiently spaced (>> $30$ s) if treated as separate data points.

4 Behaviour Sampling

• Watch entire group; record every occurrence of a specified rare but important act (e.g. fights, copulations).
• Provides full census of low-frequency events that focal/scan would miss.
• Shares conspicuousness bias with scan sampling; sometimes called “conspicuous behaviour recording.”

Recording Rules ( “How” )

Two master types (see Fig $5.1$ hierarchy):

A Continuous Recording (CR, “All-Occurrences”)

• Document exact start/stop times (states) or occurrence times (events).
• Yields true frequencies, latencies, durations, and permits sequential analyses.
• Bias risk: truncation if bout runs beyond observation period or visibility window → longer bouts more likely under-estimated.
• Labour-intensive; limits categories that can be monitored concurrently.

B Time Sampling

• Observation divided into successive sample intervals; cue at each sample point (end of interval).
• Less information retained, but multiple categories feasible.
• Two principal sub-types:

1. Instantaneous Sampling (IS) – record whether behaviour is occurring at the sample point.
2. One–Zero Sampling (1/0) – record whether behaviour occurred at any time during the preceding sample interval.
   • Both produce a single dimensionless session score:
   – IS score = \dfrac{\text{# points behaviour present}}{\text{total sample points}}.
   – 1/0 score = \dfrac{\text{# intervals behaviour occurred}}{\text{total intervals}}.
   • Individual points/intervals within a session are not independent datapoints.

Instantaneous Sampling – Key Traits

• Alternate names: scan sampling, point sampling, fixed-interval time-point sampling.
• Appropriate for clearly definable states (posture, proximity, activity).
• Inappropriate for discrete, short events or very rare acts.
• Observer must guard against “window-creep” (stretching instant to visible window, inflating conspicuous acts).

One–Zero Sampling – Key Traits

• Also “fixed-interval time-span sampling.”
• Records presence/absence within interval, ignoring frequency & duration.
• Critics: some argue for abandonment; Zinner et al. $1997$ + empirical work show acceptability with small intervals.
• Especially useful for rapid, intermittent behaviours that start/stop quickly (e.g. mammalian or avian play; Diamond & Bond $2004$ on NZ parrots).
• Systematic biases:
– Over-estimates duration (treats any occurrence as full-interval).
– Under-estimates bout count (multiple occurrences in same interval collapsed).
– Error sensitive to ratio mean bout length : interval length.

Choosing an Appropriate Sample Interval

• Trade-off:
– Shorter interval → higher accuracy, lower observer capacity.
– Longer interval → easier reliability, lower precision.
• Practical laboratory range: $10$–$60$ s (common $15$, $20$, $30$ s). Fieldwork or long sessions often need longer intervals.
• Objective determination (requires large CR baseline):

1. Obtain true record via continuous recording.
2. Simulate IS/1-0 with candidate intervals (e.g., $10,20,30$ s …).
3. Compute discrepancy vs. CR; find “break-point” where error exceeds pre-set tolerance (e.g., $\le$ $10\%$).
4. Select longest interval below break-point that satisfies all behaviour categories.
   • Caveat: impossible if CR is already impracticable—the common reason for adopting time sampling.

Comparative Worked Example (Fig $5.5$)

Assume $16$ sample intervals, each $10$ time-units (total $160$ units). Four bouts (durations $a$, $b$, $c$, $d$) with $a=39$, $b=16$, $c=12$, $d=28$ units.
• Continuous Recording:
– Total duration = $95$ units; mean bout duration = $\dfrac{95}{4}=23.8$ units.
– Proportion of time = $\dfrac{95}{160}=0.59$.
– Bout frequency = $\dfrac{4}{160}=0.025$ per unit time.
• Instantaneous Sampling: $\dfrac{9}{16}=0.56$ → close to true duration proportion, captures correct bout count.
• One–Zero Sampling: $\dfrac{13}{16}=0.81$ → substantial over-estimate of duration, records only $3$ bouts.

Advantages & Disadvantages Summary

• Continuous Recording: maximal detail for frequency, duration, latency, sequences; manpower-heavy; fewer categories trackable; susceptible to truncation bias.
• Instantaneous Sampling: lower effort, allows many categories/subjects, near-true duration if interval ≪ bout length, but misses short/rare events & sequences; vulnerable to observer bias at sample point.
• One–Zero Sampling: sole practical option for rapid on–off acts; composite index strongly correlated with frequency & duration; systematically biases absolute estimates; error varies with bout length.
• General Time Sampling Benefits:
– Reduces workload → more categories & subjects feasible (e.g., $12$ animals scanned cyclically every $15$ s → each seen every $3$ min).
– Often higher inter-observer reliability due to simplicity.
– Necessary when behaviour occurs too fast for CR.
• General Time Sampling Costs: information loss; inaccurate frequencies/durations unless interval kept short; limited for sequence analysis.

Practical & Ethical/Philosophical Implications

• Methodological rigour demands explicit protocols for visibility loss, interval timing, order of focal subjects, spacing of scans.
• Bias awareness critical: conspicuousness, privacy-seeking during out-of-view periods, and observer expectancy can skew datasets.
• Ethical responsibility: choosing least intrusive method that still yields valid inference; acknowledge limitations transparently in publication.
• Philosophically, measurement choices shape “reality” of behavioural data—researcher must match method to research question while minimising artefacts.

Key Take-Home Guidelines

• Separate which/when (sampling rule) from how (recording rule).
• Prefer continuous recording when exact metrics or sequence analyses are essential & feasible.
• Use instantaneous sampling for moderate/long states across many individuals.
• Use one–zero sampling for rapid, repetitive acts where other techniques fail.
• Select the shortest practicable sample interval that preserves reliability after training.
• Always document decisions on pursuing lost focal subjects, scan spacing, and interval length selection.
• Where possible, validate time-sampled data against continuous baselines or check correlation with frequency/duration measures.