Observational Study of Behavior: Sampling Methods

1. What they did

Whenever an observer cannot watch and record everything every animal in a group does, they must choose what to record and when. Jeanne Altmann points out that most observational studies of the time gave only partial descriptions of how they sampled behaviour, and seldom explained why they chose their method. Often the observer did not even realise a choice was being made, yet the choice decides which questions the data can answer.

This paper is an observer’s guide. It describes every method then known for sampling spontaneous social behaviour by direct observation in groups of people or other animals, and sets out the uses and limits of each. It does not cover choosing research questions or defining behaviours; it assumes the observer already has a clear question and a list of behaviours to record.

She first explains the basic choices behind any sampling plan:

• Events or states. Events are moments (an animal sits down); states have a duration (an animal is sitting). Questions about how often something happens need events; questions about how long or what share of the time need states.
• Frequency or rate. A count only means something if observation time is known; many comparisons in the literature are really comparisons of rates.
• What to record. Who acted, who received the act, the order of events, times, durations, and context such as habitat, weather and neighbours.
• When to start and stop sessions. By a schedule, at random, or triggered by a behaviour, and how unplanned starts and stops can bias the data.
• How to choose individuals. At random, by a set rotation, or by behaviour (such as the first pair to interact), which can link the sample to the behaviour itself.

She then reviews seven sampling methods in turn, with examples from published studies, and ends with a summary table of what each method is good for.

2. What they found

The seven methods and their recommended uses (from the summary table)

• Ad libitum sampling (no system; “typical field notes”): mainly useful for ideas and for recording rare but significant events. Not suitable for comparisons, because the observer’s attention is drawn to conspicuous behaviours and individuals.
• Sociometric matrix completion (extra watching of particular pairs to fill in a table of who did what to whom): suited only to asking how one-sided the relationship is within each pair.
• Focal-animal sampling (all actions of one individual, or a small group, recorded for a set time, along with how long the individual was in view): suited to sequences, percent of time, rates, durations and nearest-neighbour relationships. Usually the method of choice.
• All occurrences of some behaviours (every instance of a conspicuous behaviour in the whole group): suited to synchrony and rates, but only when viewing is excellent and the behaviour is attention-catching and not too frequent.
• Sequence sampling (following an interaction sequence from start to finish, whoever is involved): suited to the order of events within interactions.
• One-zero sampling (scoring only whether a behaviour happened at least once in each short interval): no recommended use.
• Instantaneous and scan sampling (recording what each individual is doing at preset moments): suited to percent of time, synchrony and subgroups.

Ad libitum sampling and correction factors

• Some individuals are seen more easily than others because of size, activity or position. Several researchers tried to correct for this with separate “observability” counts.
• Altmann shows that these corrections rest on assumptions that had never been tested. A shy, subordinate male might mate out of sight; a chase brings both animals into view; and an observer may unconsciously pay more attention to rarely seen animals. A correction can then make the bias worse rather than better.

Focal-animal sampling in practice

• Following one focal individual is usually best when conditions are less than perfect. If one of three focal animals moves out of sight, following it means losing the other two.
• Observer fatigue sets a limit on session length. In the Altmanns’ baboon study, even with two observers, one 15-minute focal sample per hour was close to the upper limit when recording about five dozen social behaviours. In a later study they used 40-minute samples, then rested and swapped roles.
• Time out of sight should be recorded, for example with a cumulative stopwatch.

One-zero sampling

• One-zero scores were widely used, especially in studies of children and caged monkeys, and were often treated as frequencies or as percent of time. Altmann shows they measure neither: they count intervals, not acts, and the result mixes duration and frequency in an unknown way.
• The two common justifications, that scoring is easier and that observers agree more, do not hold. Instantaneous sampling is at least as easy, and observers can agree closely on scores that are still inaccurate.

Instantaneous and scan sampling

• Scans should be as quick and even as possible, or they become a series of short focal samples of unknown length. Reported scan times ranged from 5 to 10 seconds for subgroups of children to three to seven minutes for subgroups of up to 331 individuals in gelada herds of up to 425; scans of about 40 baboons took 45 to 60 seconds.
• These samples give good estimates of percent of time, but give no information on rates of events unless further assumptions are made.

3. What they concluded

• The key question in choosing a sampling method is whether it gives a biased sample of the behaviours being studied. Some methods are biased for many variables, others for fewer. Bias can be direct, or indirect when the variable studied is linked to one that is biased.
• The observer needs to think carefully about the animals, their social relationships and the research question before choosing a method, and should avoid built-in biases as far as possible, especially those that directly affect the variables under study.
• More than one sampling method can often be used in a study, one after another or, in good conditions, at the same time. For example, during focal-animal samples the observers also recorded the neighbours of the focal individual every five or ten minutes, and all occurrences of conspicuous group behaviours such as predation, contact between groups and drinking.
• Sampling decisions give field workers a way to raise the internal validity of observational studies without manipulating the animals, so field studies can be as rigorous as laboratory work while staying relevant to real-world behaviour.

6. Other technologies, skills or innovations

• A cumulative stopwatch to record the time a focal animal is in view; a stopwatch with an extra “marker” hand for recording behaviour transitions.
• Two-observer teams: one watches the focal animal and dictates, the other records (in Slatkin’s baboon study, the assistant also drove the vehicle).
• Tables of random numbers for choosing focal individuals and session times.