The SI system of units (Système International d'Unités) should be used for measurements. The SI system is completely coherent, which means that all derived units are formed by simple multiplication or division of base units without the need for any numerical factors or powers of ten. This distinguishes the SI system from earlier metric systems such as the centimetre–gramme–second (CGS) system, which it superseded. The SI system comprises nine base units, each of which is independently defined, and various other units which are derived by combining two or more base units. The base units, together with some of the more common derived units, are listed in Table A1.1. Some common non-SI units and their SI equivalents are shown in Table A1.2.

Conventions. Each unit is represented by a standard unit symbol (e.g. m, s, A, kg), which may be multiplied or divided by other unit symbols or numbers (e.g. 3 m, 0.12 kg m, 16.5 m s−2). Unit symbols are algebraic symbols and follow the conventions of algebra. They are not abbreviations, and should never be followed by a full stop or an ‘s’ (to denote plural). The names of units (e.g. metre, second, ampere) are all spelt with a lower case initial letter. Symbols for units named after a person start with an upper case letter (e.g. A for ampere, K for kelvin, Pa for pascal).

In this chapter we consider some particular aspects of the analysis of behavioural data. We begin with the crucial dimension of time. We consider a number of ways in which order can be extracted from the observed stream of behaviour. We go on to consider how best to treat the data obtained from choice tests as described in Chapter 8 and conclude with some ways of dealing with social behaviour.

Bout length

An estimate of bout length may be required when behavioural acts recur in temporal clusters (a bout of events) or when the same, relatively prolonged behaviour pattern occurs continuously for a period (a bout of a single behavioural state). If behaviour patterns are neatly clumped into discrete bouts separated by uninterrupted gaps, then one bout can be distinguished from the next with relative ease. Often, though, bouts are not obviously discrete, in which case a statistical criterion must be used to define a single bout of behaviour. One commonly used technique is log survivorship analysis. This is a simple graphical method for specifying objectively the minimum interval separating successive bouts: the bout criterion interval (BCI). Any gap between successive occurrences of the behaviour that is less than the BCI in length is treated as a within-bout interval, while all gaps greater than the BCI are treated as between-bout intervals. To estimate the BCI, the cumulative frequency of gap lengths (on a logarithmic scale) is plotted against gap length (on a linear scale).

Many students are given ready-made problems on which to work but it pays to think carefully before you start a project, whatever stage you are at in your scientific career. Sage advice is given in the book by Cohen and Medley (2000). Here we are concerned with the particular issues that need prior thought in behavioural biology and psychology.

Choosing the level of analysis

Behaviour can be analysed at many different levels, from the complex social interactions within populations to the fine spatial detail of an individual organism's movements. A simple but fundamental point is that the form of measurement used for studying behaviour should reflect the nature of the problem and the questions posed. Conversely, the sorts of phenomena that are uncovered by a behavioural study will inevitably reflect the methods used.

A fine-grained analysis is only appropriate for answering some sorts of question, and a full understanding will not necessarily emerge from describing and analysing behaviour at the most detailed level. While a microscope is an invaluable tool, in some circumstances it would be useless – say, for reading a novel. In other words, the cost of gaining detail can be that higher-level patterns, which may be the most important or relevant features, are lost from view. For example, recording the precise three-dimensional pattern of movements for each limb may be desirable for certain purposes, such as analysing the neurophysiological mechanisms underlying a particular locomotor behaviour pattern.

We are pleased that many of the issues that were relatively novel in behavioural biology when we wrote the first edition (1986) of this book have now passed into the mainstream of methodological thought. Nevertheless, we believe that the principles are worth reinforcing.

In this edition we have changed the structure so that greater prominence is given to the non-experimental aspects of behavioural biology. Some behavioural research simply involves carefully watching an animal to see what it does next. Performing an experiment may seem more ‘scientific’ than open-ended observation but the yield may be less. Moreover, worthwhile experimental research almost invariably needs to be preceded by careful observation. Knowledge of the normal behaviour of animals, preferably in their natural environment, is an invaluable precursor to experimental research.

We have also expanded the section on research design because, more than ever, good design can make such a difference to how big the sample must be, the interpretation of data and the time taken to prepare results for presentation or publication when the moment arrives. We have eliminated the further reading sections at the end of each chapter, but have given advice on further reading at appropriate places in the chapters. Each chapter now ends with a summary. We have taken out the annotated bibliography that formed such a large part of the reference section in the two previous editions (1986 and 1993) because we felt that such material was not essential to the main purpose of the book.

Having decided which aspects of behaviour to measure and chosen the suitable recording medium, you would be wise to check the quality of your measurements before you proceed to collect a lot of data. Measuring behaviour, like measuring anything else, can be done well or badly. When assessing how well behaviour is measured, two basic issues must be considered: reliability and validity – sometimes expressed as the distinction between ‘good’ measures and ‘right’ measures.

Reliability versus validity

Reliability concerns the extent to which measurement is repeatable and consistent: that is, free from random errors. An unbiased measurement consists of two parts: a systematic component, representing the true value of the variable, and a random component arising from imperfections in the measurement process. The smaller the error component, the more reliable the measurement. Reliable or good measures are those that measure a variable precisely and consistently.

At least four related factors determine how ‘good’ a measure is:

• Precision: How free are the measurements from random errors? The degree of precision is represented by the number of significant figures in the measurement. Note that precision and accuracy are not synonymous: accuracy concerns systematic error (bias) and can therefore be regarded as an aspect of validity (see below). A clock may tell the time with great precision (to within a millisecond), yet be inaccurate because it is set to the wrong time.

• Sensitivity: Are small changes in the true value invariably reflected by changes in the measured value?

• Resolution: What is the smallest change in the true value that can be detected?

• […]

Data can always be interpreted in a variety of ways. As you think about your findings, you would be wise to be sceptical, at least initially, about your own preferred explanation. Colleagues can usually be counted on to help you in considering alternative ways of accounting for your results. Many of the problems that arise in the interpretation of data can, however, be avoided if the research is carefully designed. We have discussed some of these issues in Chapters 7 and 8 – for example: are the measurements independent of each other? Is generalisation to another group of subjects limited because of the design? Have the results been affected by the order in which the treatments were presented? Did the observer influence the subjects in some way? Did you unwittingly select the data that fitted a preconceived idea or bias?

In well-designed research these issues will have been considered in advance and appropriate precautions taken. However, problems of interpretation that could have been foreseen often arise through oversight or the sheer practical difficulties of avoiding them. It is often hard, for example, to conduct an experiment ‘blind’. The solution is not to ignore the problem, but rather to acknowledge it. You should be alert to potential difficulties and be honest about how these possibilities might affect interpretation. In this chapter we mention some additional issues that might affect how data are interpreted and discuss the presentation of results to a wider audience.

The options available

Having looked at the various forms of behavioural measure, we now move on to consider the mechanical processes involved in recording them. The choice of the medium, or physical means, used to record behavioural observations has important consequences for the sorts of data that can be collected and the sampling techniques that can be used. Five basic methods of recording behaviour are available: video recording; written or dictated verbal descriptions; automatic recording devices; paper check sheets; and computer event recorders. The most flexible and commonly used methods are check sheets and computer event recorders. Using check sheets, an event recorder or any other method obviously pre-supposes that you have formulated a set of discrete behavioural categories.

Video recordings give an exact visual (and perhaps audio) record of the behaviour, which can subsequently be slowed down for analysis. Such date-stamped evidence is sometimes used as proof that the observer saw what was claimed and may be needed for inspection by others. Video recordings are useful for studying behaviour that is too fast or too complex to analyse in real time. Similarly, exact records of vocalisations can be made with an audio recorder and the sound patterns analysed later using specialised software. Digital technologies have largely superseded analogue methods. It is worth remembering, however, that when video recordings are stored on a computer they may be compressed and thereby lose quality.

The steps involved in studying behaviour

Recipes for conducting research are rarely followed precisely and most scientists build their ways of investigation in periods of apprenticeship when they model themselves on the behaviour of more experienced colleagues. In considering the steps listed below you should be aware that many programmes of research enter the sequence at different points. In general, though, studying behaviour involves a number of inter-related processes in roughly the sequence in which we have listed them. We have described these steps in outline. Lehner (1996) and Hailman and Strier (2006) provide much fuller accounts of research methodology, although we depart from their schemes – particularly in the emphasis we have placed on our first five steps.

Ask a question

Before any scientific problem is investigated, some sort of question will have been formulated. The question may initially be a broad one, stemming from simple curiosity about a species or a general class of behaviour, such as ‘What does this animal do?’ Such a question is not a hypothesis.

The value of broad description arising from sheer curiosity should not be under-estimated. Alternatively, it may be possible at an early stage to formulate a much more specific question based on existing knowledge and theory, such as ‘Do big males of this species acquire more mates than small males?’ This is tacitly a hypothesis. Not surprisingly, research questions tend to become more specific as more is discovered about a particular issue.

This edition of Measuring Behaviour does not contain the long annotated bibliography found in the first two editions. That bibliography contained references to the most important publications relating to the development of the methodology used in the direct observation of behaviour and has historical interest. For those who would like to consult it, you can visit the annotated bibliography of the second edition on the following website: www.cus.cam.ac.uk/~ppgb/.

The brief annotated bibliography contained in this edition relates solely to statistics books that we have found useful. It is far from comprehensive but should provide a helpful entry into a large and ever-growing library of books on the subject.

General advice on statistics

We appreciate that most people who read this book will learn most about statistics when they start to analyse their own data. In general, we advise against poorly informed cook book approaches to statistical analysis. Our aim in this chapter is to consider briefly some of the main issues that arise in the statistical analysis of behavioural data. This is not a statistics textbook, however, and for an account of statistical methodologies you should consult one of the many excellent books available, but preferably one that has as its target a biological or psychological audience (e.g. Zar, 1999; Sprinthall, 2003). In Appendix 3 we have given an annotated list of books that we have encountered.

Given the inherent variability in biological systems, statistical analysis is often essential for unravelling what is going on. Nonetheless, excessively complicated statistics are sometimes used as a substitute for clarity of thought or good research design. Do not become obsessed by statistical techniques, nor too cavalier in their use. Statistical analysis, no matter how arcane or exquisite, can never replace real data.

Besides sometimes being over-used, statistical techniques are frequently misused in the behavioural literature. We have already outlined one common error in Chapter 7 – that of including many data points from the same individual in the mistaken belief that they are independent measurements. In Chapter 11 we consider, among other things, the misinterpretation of multivariate statistics and the various misuses of correlation coefficients.

Sampling rules

When deciding on systematic rules for recording behaviour, two levels of decision must be made. The first, which we refer to as sampling rules, specifies which subjects to watch and when. This covers the distinction between ad libitum sampling, focal sampling, scan sampling and behaviour sampling. The second, which we refer to as recording rules, specifies how the behaviour is recorded. This covers the distinction between continuous recording and time sampling (which, in turn, is divided into instantaneous sampling and one-zero sampling; see Fig. 5.1). Do not use ‘focal (animal) sampling’ as a synonym for continuous recording described below. To do so would conflate a sampling rule (which individual is watched) with a recording rule (how behaviour is recorded).

In this section, we consider the four different sampling rules.

Ad libitum sampling means that no systematic constraints are placed on what is recorded or when. You simply note down whatever is visible and seems relevant at the time.

Clearly, the problem with this method is that observations will be biased towards those behaviour patterns and individuals which happen to be most conspicuous. For example, ad libitum sampling tends to miss brief responses and underestimates the involvement of some age groups in social interactions (Hernández-Lloreda, 2006). Provided this important limitation is borne in mind, however, ad libitum sampling can be useful during preliminary observations, or for recording rare but important events.

Formal prescriptions about how scientific research should be conducted often fail to capture the creativity of the best scientists. Therefore, advice on how to design research must be given and taken with caution. Some research might simply involve carefully watching an animal to see what it does next. This type of work should not be scorned. Performing an experiment may seem more ‘scientific’ than open-ended observation but the yield may be less. Many questions about behaviour are most appropriately answered by non-experimental observational research. Such work can also help to distinguish between alternative explanations if, for example, naturally occurring events demonstrate associations between variables that previously seemed unrelated, or break associations between variables that previously seemed to be bound together. Moreover, worthwhile experimental research almost invariably needs to be preceded by careful observation. Knowledge of the normal behaviour of animals, preferably in their natural environment, is an invaluable precursor to experimental research.

Performing experiments

The point of an experiment is to find out whether varying one condition produces a particular outcome, thereby reducing the number of plausible alternative hypotheses that could be used to account for the results. You will almost inevitably have some expectations about the outcome of an experiment, even if you are not consciously aware of these expectations. This potential source of bias can be removed by ensuring that the person making the measurements is unaware of which treatment each subject has received until after the experiment is over.

The scope of this book

This book is intended as a guide to all those who are about to start work involving the measurement of directly observed behaviour. We hope it will also be useful for those wanting to refresh their memories about both the possibilities and the shortcomings of available techniques.

Those who have never attempted to measure behaviour may suppose from the safety of an armchair that the job is an easy and straightforward one, requiring no special knowledge or skills. Is it not simply a matter of writing down what happens? In sharp contrast, those attempting to make systematic measurements of behaviour for the first time are often appalled by the apparent difficulty of the job facing them. How will they ever notice, let alone record accurately and systematically, all that is happening? The truth is that measuring behaviour is a skill, but not one that is especially difficult to master, given some basic knowledge and an awareness of the possible pitfalls.

The purpose of this book is to provide the basic knowledge in a succinct and easily understood form, enabling the beginner to start measuring behaviour accurately and reliably. A great deal of high-quality behavioural research can be done without the need for specialised skills or elaborate and expensive equipment.

Sometimes it is possible to carry out behavioural research simply by relying on written descriptions of what the subjects do. Usually, though, worthwhile research will require that at least some aspects of the behaviour are measured.

Identifying individuals

In many studies, being able to identify individuals is essential. Focusing on the behaviour of an individual in a group is virtually impossible without a way of distinguishing reliably between one individual and another. Moreover, when differences in behaviour of known individuals are recorded, the resulting data are likely to be much more informative. Only by identifying and watching individuals does it become clear that all individuals in a species do not behave in the same ‘species-typical’ way.

In the laboratory, identification of individuals by rings, tags, collars, tattoo marks, painting the skin, dying feathers, fur-clipping and so forth does not usually offer major practical difficulties. However, it is important to realise that marking an individual may alter its behaviour or that of other individuals with which it interacts. To give one example, research revealed that coloured plastic leg bands placed on male zebra finches affected how attractive they were to members of the opposite sex. Female zebra finches preferred males wearing red leg bands over unbanded males, while males preferred females with black leg bands. Both males and females tended to avoid members of the opposite sex wearing green or blue leg bands (Burley, 2006). These findings clearly show that for zebra finches, and probably many other species, methods conventionally used to identify them can have a significant effect on behaviour.