SYSTEMATIC OBSERVATION OF BEHAVIOUR

 

SYSTEMATIC OBSERVATION OF BEHAVIOUR

 

© Jacques P. BEAUGRAND 

beaugrand.jacques@uqam.ca

Version 3.1

 

Plan of this text

ABSENCE OF UNEQUIVOCAL TERMINOLOGY

BEHAVIOUR: A VERY BROAD CONCEPT

CRITERIA OF DESCRIPTION AND CLASSIFICATION

A. Concrete criteria 

a) Classification according to form

b) Classification according to effects

B. Theoretical and abstract criteria

a) Classification according to cause

b) Classification according to function

 

CHOICE AND DEFINITION OF UNITS

A. Choice of units

B. Definition of Units

C. Repetitions and transitions

 

THE DESIGN OF OBSERVATION

A. The choice of taxonomy

B. Recording of temporal elements

C. Methods of recording

D. Encoding and Notation Systems

E. Complete and continuous observation as opposed to sampling

F. Back-up techniques

G. The degree of generalisation

H. Factors to be controlled

a) Effects due to observer interference

b) Expectations of subjects or observers

c) Instrumental decay

d) Example

 

SAMPLING TECHNIQUES

A. Non-structured or ad libitum sampling

B. Complete and continuous sampling

C. Sampling by successive focalisation

D. Sampling in Sequence

E. Sampling by presence or absence

F. Sampling by instantaneous scanning

G. Matrix completion

H. Choice of a technique

I. Instrumental accuracy

 

 

APPENDIX

Agonistic and epigamic behaviours of Xiphophorus helleri

 

REFERENCES

    The objective of the present chronicle is to serve as a handy reference for the reader about the difficulties and techniques involved in the direct observation of behaviour. These techniques may be defined as bringing about observations of behaviour, without the use of instruments which express, in a physical form, what occurs during observation. Scales, thermometers, levers, electrocardiographs, even surveys or computers programmed to recognise formal patterns, are instruments which make the recording of results less debatable, less subject to fluctuations in perceptive, opinion, conceptual or theoretical differences between human observers. Technology and theoretical knowledge are not always able to provide us with instruments adequate to our needs. Research must therefore advance by the use of direct observation, which is to say, by the use of humans as instruments who produce observations directly from the information received by their senses and interpreted with reference to preconceptions. Direct observation of behaviour consists in identification, naming, comparison, and finally, description of the behaviour itself. A thorough command of sampling techniques is of course important. However, the techniques of direct observation of behaviour, are of only a secondary use if the observational units and exploratory patterns have not first been clearly and carefully defined so as to form modal regularities that the human observer may reliably recognise and form a system of recognition.

 

ABSENCE OF UNEQUIVOCAL TERMINOLOGY

    The absence of an unequivocal terminology engenders three common errors when the direct observation of behaviour comes into question. The first opposes the experimental method and the approach to phenomena by direct observation.

It is possible, and frequent, that direct observation is used in the context of experiments upon one or many independent variables. For example, to determine the effects of population density upon social behaviour in broiler chickens, we could create equivalent groups of individuals which would be placed in enclosures. The dimensions of each enclosure would be experimentally manipulated. The research plan would co-ordinate both the independent variable and the method of observation. The different behavioural patterns to be reported would therefore constitute as many dependent variables. This situation, would differ from the more common experimental situation in which few dependent variables are recorded.

    There is often also a tendency to confuse research which uses direct observation with research that proceeds without hypotheses. First of all, we have just mentioned that direct observation could be used in experimental research. Experimental research necessarily formulates hypotheses concerning the possible relation between one or more independent variables and one or several dependent variables. At least such hypotheses are here implicit in the research design. Moreover, it is absolutely possible to state hypotheses in the context of non-experimental research, in which behaviour is directly observed. These hypotheses may concern the effects of naturally-occurring variations, be they in the environment or in the individuals studied. In systematically recording observations under both or several of these conditions, or in selecting those observations concomitant to one and not to others, in order to compare them, we perform what Tinbergen (1953) called natural experiments. Climatic and seasonal variables, the type of environment, the personal characteristics of individuals, and other variables related to the subjects may be selected so that concomitant behaviour may be compared. These kinds of approaches are also called quasi-experimental, regressive and ex post facto studies by behavioural scientists (Kerlinger, 1973).

    Third, descriptive hypotheses stated by the researcher may also concern the existence of regularities in the facts, or in the relations between them. For example, evolutionary game theory may predict the existence of a much-ritualised challenging behaviour in the repertoire of a species which possess dangerous arms such as antlers. Direct observation may provide data to verify or not the existence of such a regularity and give indirect support to the theory.

    Finally, direct observation of behaviour is often confused with naturalistic or ethological observation. Although technically they share essentials, the preoccupations are quite different. Thus, it is possible to perform direct observation in the absence of any ethological preoccupation. Direct observation may be used in naturalistic observations, in which the observer studies behaviour with as little interference as possible, the essential concern being respect for the environmental conditions, be they physical or social, which have biological or adaptive significance for the subject. But, it is also possible to study human and animal behaviour through direct observation in highly artificial environments, such as battery cages, prison, or orbital station. Although far from being natural, these situations are still highly significant for men and animals. Their animal repertoires do not seem to be considerably modified in "unnatural" environments.

 

BEHAVIOUR: A VERY BROAD CONCEPT BUT SPECIAL CONCEPT

    In order to directly observe behaviour, it is necessary to establish a notion of behaviour which is broad enough to be useful, while be checkable, testable, or to be part of verification. We shall adopt a realistic conception of behaviour. Behaviour is basically a biological production, it is basically an observable or obectivable spatiotemporal modification within the form or complexion of the behaving organism. Behaviour is organized hierarchically within the individual. Behavioural levels emerge from social and populationist interactions. Let us examine each of these properties in turn.

 

    Basic behaviour is biological:  a behaviour is produced by one individual, or by two individuals interacting, or even by a set of connected living organisms which influence each others. By influence it is meant that there is some kind of connection between individuals, and at least one of the individuals behaves differently from the way it would behave when alone. We shall also agree to consider as behaviour some physiological activities of individual nervous systems, including some very specific activities such as electromagnetic discharge of organs, the production of sound, pheromones and changes in colour (such as blushing in humans). 

 

    Basic behaviour is overtly observable:  basic behaviour must bring about observable spatiotemporal modifications which must be noticeable, perceivable by other organisms, including the human observer. 

 

    Basic behaviour is objectivable:  what is noticed or perceived by one trained observer must also be noticeable or observable by another skilled but independent observer. Validity at the level of observation comes from intersubjective agreement, i.e., two or more observers admitting observing the same thing at the same time. 

 

    Behaviour is organized hierarchically:  structuration occurs on two aspects which are not necessarily independent. A first aspect is material / biological: contraction of individual muscles contribute to the moving or change in form of individual body parts, which in turns contribute to a more complex movement such as putting a foot in front of the other, which contribute to walking, which in turn contribute to an even more complex act, such as following a mate, &c. 

    A second aspect of hierarchical organisation is essentially conceptual or taxonomical: it concerns the way the observer organises in a systemic way the various spatiotemporal modifications he or she observes. Thus, basic behaviour units which are observable can be regrouped to form more general behaviour categories, which in turn can be regrouped to contribute to even more global categories. For one example of conceptual organization of agonistic behaviour in a fish, examine Figure 1. The organisation of behaviour into a system can contribute  to emergent properties across ontological levels:  basic units of behaviour are properties of the individual organisms. It is the individual which behaves. When two or more individuals interact, new behaviours with new emergent properties can be recognized. For instance, one individual Canada goose flies, geese of the same family fly forming "à la queue leu-leu", with parents leading and yearlings following; several families originating from the same area will form a dashed line and, if their number increases, a two-dimensional V flight will get organized. At the level of the whole population several parallel V flights will progress together side by side without merging. Thus, at each of these levels, new characteristics appear for the observer and some of these were not present at the more primitive levels. The problem is that all these are considered as "behaviours". 

 

    Thus, behaviour may be studied with reference to diverse levels of organization which range from the finest (individual) to the most molar (population).  The scale, starting from the bases of neurology, physiology and anatomy get organised into fine motor patterns, moves on to comprehend the motor pattern taken as a whole, the individual acts, which are the co-ordinated motor functions related to several motor patterns, the individual transactions with the physical and social environment, the social interactions between two or many individuals, the behaviour of groups in their entirety (for example, hierarchical organization), the behaviour of entire populations and species (such as the reproductive strategy of a species) and finally, the behaviour of groups of species, families, and genera (for example, modes of reproduction in social insects).

 

    The concept of behaviour is therefore already vast enough as defined here to exclude covert behaviour such as motivation and emotion, thinking, consciousness, learning. These can be studied as brainy processes which can have some effect on behaviour, but should not be considered as behaviour in themselves. They are processes identical to or emerging from nervous activity, just as overt behaviour, but they should be kept distinct from the latter.

 

As one will easily admit, a systematic theory of behaviour organisation is badly needed: though behaviour is recognized as being the privileged object of study of several behavioural sciences, including ethology, behavioural ecology, sociobiology, psychology, education, etc, none of these disciplines has taken the time to clearly state what "behaviour" for them meant or referred to. But one thing remains essential for the pursue of direct observation of behaviour: primitives on which behaviour is constructed by the observer must be of the domain of the observable.

 

CRITERIA OF DESCRIPTION AND CLASSIFICATION

    The behaviour of an organism in interaction with its environment lies within a very complex system of events. The researcher should first describe the pertinent properties upon which will rest the recognition of behavioural units. 

    By convention, a behavioural unit will refer to one of the atoms or primitives upon which other more molar behaviours may be later constructed.

To observe is to abstract (Hinde, 1970), and also to classify. Classification made at the time of observation must be distinguished from that done after-the-fact, when the researcher, by grouping together his observations, extracts pertinent data comparable to that implied by his research hypotheses. In this latter case, the researcher uses more comprehensive classifications, and therefore more interpretative or theoretical ones, which may have been defined a posteriori; the classification is therefore provisional and modifiable. If, on the other hand, the classification is determined a priori, it is irremediably definitive, unless the researcher conserved video tape or film archives which permit repeated viewing and description of events. A priori classification may be made from concrete or descriptive criteria, or from theoretical or interpretative criteria.

A. Concrete criteria

    Concrete criteria are related to formal, topological or kinetic properties of behaviour, or to their effects upon the environment.

a) Classification according to form

 

    An act or a posture may be described according to its form or according to its formal or kinetic characteristics. Formal characteristics are identified according to the muscular contractions involved, as well as by the position of limbs or other anatomical, visual or auditory characteristics. For example, bird song may be recognised by its tonal characteristics. Most postures, gestures, motor patterns, facial expressions and specific vocalisations may be described according to such criteria. Usually, a general reference is made to the organization or to the characteristics of the form in question. Effectively, it is not always necessary to name all muscles or anatomical structures involved, nor their specific state of more or less partial contraction, unless the study is oriented specifically to a consideration of these details.

    Use of formal criteria can be very laborious if the behaviour studied is rapid and complex. The description of behaviour in such fine detail is useless in most cases, as the construction and use of more comprehensive behavioural units more intelligibly takes its place. The level of synthesis or detail employed will of course depend upon the researcher's interests, and on the orientation of the research. Although formal criteria may be difficult to apply when recording behaviour by notation, these criteria are nevertheless essential to visual recognition of the basic patterns which make up more easily recorded comprehensive behavioural units.

    Each behavioural pattern may be characterised by a number of dimensions which Drummond (1981) identifies, in a rather arbitrary manner, as spatial location, orientation, concrete topography and intrinsic properties. He suggests initial examination of these dimensions, which may reveal formal regularities. Static conditions or changes in one or more dimensions will supply criteria for recognition of behavioural patterns, which will be in turn described, named and defined.

First, spatial location concerns the place in which the individual is found, the place he goes to, and the place over which his trajectory lies. This location is defined in relation to elements of the environment. Therefore, if an individual charges, approaches, climbs up a tree or broods, these actions are all performed with reference to a place. The first three actions involve more or less rapid changes in locations; the last implies, to the contrary, a consistency in location, defined in relation to the nest or its contents.

    Second, orientation concerns the positions of the individual's anatomical parts as they related to other forms, organic or otherwise, present in the environment. This may involve the direction of anatomical parts toward objects, as in pointing of looking at. For example, in the fish Lebistes reticulatus, two motor patterns, which are formally identical, may be classified differently due to their orientation. The first is performed parallel to the female (parallel gallop and parallel lateral swim) whereas the second is performed directly in front or her, and blocks her way (blocking).

Third, three-dimensional topography of the individual is an important source of regularities, permitting characterisation and recognition of motor patterns. These regularities may be recorded in the movements of certain members (legs or wings), and anatomical parts (crests, fins, tails, or brows) as well as in the static arrangement of these structures (swelling of certain structures, piloerection, spreading of wings or raising of eyebrows). These regularities are noted by referring to the individual anatomical structure, or to other structures.

Finally, many organisms are able to modify certain intrinsic properties of their bodies and skin. For instance, certain animals can change colour, or body temperature, or their capacity to reflect incident light may vary. When these states are observable, they may help to make essential elements of behavioural patterns recognisable.

b) Classification according to effects

    A description may also be made according to the physical effects of behaviour, which affect conspecifics or objects present in the environment; spatiotemporal consequences are also noted. Many types of behaviour have direct effects on the environment: they move, deform, separate, consume, destroy or create components of the environment. Physical changes of a mechanical, chemical or electrical origin enter in this category, but not the reactions of conspecifics. The physical effect of a certain behaviour is only one aspect of this classification; dejections, deposits, emission of sounds, odours or venom also fall into this category.

Although it is more synthetic, this method of description according to physical effect has many advantages. One expression or term can effectively summarise a description which would otherwise involve a description of multiple muscular contractions in proper sequence. Also, one term may designate many types of actions, or motor patterns, which may themselves be described according to their consequences. A hierarchical linguistics of behaviour may be determined. For instance, "attack" may group together the actions of running, swimming, flying or walking towards a conspecific, or may consist of a rapid movement toward him, if this movement is followed by biting or striking. Behavioural units may also be defined objectively in terms which only involve changes in the physical environment. So, behaviour such as eating or mating have very immediate consequences which are difficult to contest. The term "eating" includes the act of bringing food to the mouth, and also includes the fact that the food disappears. This definition does not imply anything but the description of what the individual does, regardless of the underlying controlling mechanism, or locomotive or other details which bring about this effect. Finally, description according to the objective consequences of actions makes it possible to include, in the description, the relations between the initiating individual and his spatiotemporal context, involving location, objects, conspecifics and other anterior or posterior behaviour. This makes it easier to establish regularities. On the other hand, this method of description has its disadvantages. First of all, details are lost. If postures and motor patterns are described at a relatively molar level, the information relating to the particulars, of which behavioural units are composed, is sacrificed. In the second place, there is a loss of information as to what temporally follows a behaviour which might be a part of a chain of postures and motor patterns, or which might be the consequence of such a chain. For instance, when a child quickly approaches another, does it constitute an attack or flight from a third person? Finally, there is a serious risk of overinterpretation, for the description of behaviour according to its immediate effects easily leads to a description of adaptive consequences or of the function of the behaviour.

B. Theoretical and abstract criteria

Theoretical and abstract criteria are interpretations of certain concrete clues, using postulates, hypotheses or even theories which are not necessarily and explicitly clear to the observer. These criteria lead, for example, to causal, functional or teleological classifications. Under this same heading, we may include classifications which depend upon the intentions accorded to subjects.

 

a) Classification according to cause

Different types of behaviour may be grouped together under the same heading when they have the same cause, in as much as the factors which seem responsible for their appearance are judged to be the same. These factors include common physiological and neurological mechanisms and processes, as well as common exogenous stimuli. To illustrate: behaviour which varies in frequency and intensity in relation to changes in sex hormones are liable to be put into the category of sexual behaviour, whereas those brought about by the presence of young are likely to fall into the category of maternal or parental behaviour. Likewise, all activities influenced or brought about by the presence of a male rival may be classified as agonistic behaviour. This method of classification is essential to the understanding and explanation of behaviour. In general use, this method nevertheless poses some problems related to ethological, causal analysis, among others, that of first showing that a valid identification of causes has been made, which is more easily said than done.

b) Classification according to function

    Behaviour can also be classified according to a presumably common function. For example, all behaviour used to obtain food is grouped together under the heading of feeding behaviour, while that leading, on a more or less long-term basis, to individual reproduction falls into the category of reproductive behaviour. Epigamic behaviour has the function of synchronising two sexual partners. Likewise, that leading to the establishment of dominance or hierarchical submission belongs to the category of agonistic behaviour.

    In certain cases, it is possible to experimentally establish the function of a particular behaviour. Pruscha and Maurus (1975) were able to classify social signals (calls and postures) produced by the squirrel monkey (Saimiri sciureus) according to their immediate effects upon the behaviour of individuals which were stimulated, by telemetry, in the specific areas of the brain controlling the emission. In the absence of experimental demonstration, however, classifying behaviour according to its presumed function carries with it a good deal of speculation.

Practically, if we consider a single animal species, the causal and functional classifications overlap considerably. This is normal within certain limits, since, from an evolutionary point of view, the underlying mechanisms of adapted behaviour should be much less complex if functionally linked activities also have the same causes, than if each one should have different causes (Hinde, 1970). Therefore, certain functional categories, such as reproductive and sexual behaviour, designate causal categories as well. In most cases, classification by this criterion is facilitated by the fact that different behaviour with the same function has the same physical effects on the environment.

CHOICE AND DEFINITION OF UNITS

    Before undertaking observation itself, special attention must be paid to the choice of behavioural units which are to be recognised and noted, as well as to the definition of these units.

Choice of units

Six conditions should be respected when choosing units, so that they may be accepted according to concrete or abstract criteria, or both. First of all, the units should be discrete and exclusive. All behaviour belonging to a unit or category must share certain properties which very clearly distinguish it from behaviour belonging to other units. Second, each unit must form a homogenous category. In creating a unit, the researcher should have good reason to believe that all behaviour to be included in that unit will be equivalent as to its form, its immediate effects, its causes or its function. If inhomogeneous behaviour is included in a category, it is quite probable that at the time of analysis, regularities will be either missing or artificial. Third, it is better to increase the number of units than to combine them. If two types of behaviour are very similar formally, but can be distinguished by at least one reliable, objective criterion, it is preferable to associate them with two different units. An analysis could reveal, after the accumulation of data, that these two units are in fact different. Otherwise, the researcher might be permitted to group them into one unit. If all behaviour has, however, been noted from the beginning under the some heading, it will not be possible to divide this category into subheadings. This principle applies particularly to an exploratory situation in which the researcher has no precise hypotheses to test. This situation should enable the emergence of regularities as soon as the data is examined (Kirk, 1968). When, however, precise hypotheses are being tested, it is not necessary to be encumbered with useless observations which could only, at best, result in further hypotheses to be tested in future research.

    Fourth, abstract classifications should be avoided, from the observational stages of a research, onward. Such classifications are those suggested by a given theory. The researcher should keep any reinterpretation of his basic observations in the light of other theories to himself. Sometimes, the very names of categories have a causal or functional connotation. That a facial expression is qualified as a smile is acceptable when describing child behaviour, even if the term "smile" has a functional connotation. On the other hand, applying the same term to another species of primates is inappropriate as long as the function of the facial expression which consists of a "silent bare-teeth display" will not be accepted as homologous and the functional equivalent of the human smile. Of course, the function of the smile in humans must first be well known. The researcher must restrict himself to the use of objective terms, which are based as much as possible on the physical configurations, forms or effects of the motor patterns or postures involved. Fifth, it is important to always define, specify , and justify each behavioural unit. There is no assurance that every researcher in the field will recognise what one researcher means for example, by "lateral display" or "charge": which of these means "attack" and which "approach"? We will presently return to the problem of defining units.

    Finally, the researcher should observe only a limited number of units, depending on the research objectives. It is not easy to work with a great number of units, even if instruments help to encode and record events, and even if the analysis of data is computer-aided. It is extremely difficult to systematically observe and take note of behaviour continuously and for long periods of time if the number of units is very high. The period of observer training will be longer, reaction times higher, errors more frequent, and more attention paid to the technical aspect of observation then to that which is being observed. As soon as the researcher reaches the stages of data reduction and data analysis, his observations will have to be regrouped into more comprehensive units, which must be created a posteriori, if he wishes to extract some regularities from his observations. In some cases, it may turn out that regrouping is most efficiently performed at the same time that observation is carried out. Moreover, the greater the number of units used, the greater (exponentially) will be the number observations required in order to respect the basic postulates of certain statistical tests. On the other hand, if the number of units is too small or if they are too diffuse, it is quite possible that the results will be vague.

    As a general rule, only those units which are directly pertinent to the research objectives should be used. All in all, it is more efficient to undertake new research in order to systematically take note of neglected units which are hypothetically pertinent than to analyse data which has obviously been acquired for another reason. Laboratory cabinets the world over are bulging with uselessly amassed data. When they are analysed after a few years, the researcher has usually forgotten the conditions, often of a rather lax nature, under which the data was accumulated, and has a tendency to over-estimate their importance.

Definition of Units

    Once chosen, a behavioural unit should be defined in order to assure both intra-observer and inter-observer reliability. Specific behaviour should be defined clearly, using precise criteria. The definition of a behavioural pattern neglects, in its formulation, a large part of the information made available by a description. The definition is limited to criteria which are necessary and sufficient in order to recognise an event and classify it within a pattern. For instance, a bite may be defined as any contact between the mouth of one individual and a conspecific. This obviously does not concern the description of a type of bite. Only the strictly necessary and sufficient elements are referred to so that the observer can recognise the large majority of individual events which we usually call biting. In other words, with the species concerned, and under the conditions studied, the application of the exploratory model defined as a bite is sufficient, the rest of the information being redundant or superfluous. Currently used in ethology, such definitions do not go beyond simple identification of behavioural patterns.

The object of a definition is to assure the reliability and consistency of the instrument used, in this case, the observer, and the reliability between observers. The definition should include at least one clear, precise criterion, in reference to which a behaviour may be declared absent or present. This criterion may be formal (topological and kinetic); it may designate the actions to be completed, or the effects to be obtained or required in order to fulfill a number of criteria relating to two or more of these actions or effects. In this case, the definition is termed operational; it does not concern any procedures of measurement. It concerns the operations to perform, and the effects to be obtained following the action of observed individuals, so that the observer may be authorised to declare that such-and-such an event belongs to such-and-such a category. 

Definition, therefore, consists of assigning a label or a term integrating many essential pieces of information. 

    The researcher must also define certain arbitrary units, taking into account, for example, the changes in the spatial position of observed individuals. So, in Beaugrand et al. (1984), we provided a unit called "change of compartment", which implied that every passage of a fish from one compartment of the aquarium to another was noted with reference to the sector of origin and that of destination. IN this type of definition were combined formal criteria, and therefore such criteria related to the topology and kinetics of the patterns as the position or state of the fins, or the trajectory of the fish, and others, linked to their effects, such as mouth-contact or rapid distancing. Often, contextual elements are also introduced as essential criteria. So, the presence (preliminary or concomitant) of a conspecific in proximity, or even in a particular posture, is required in the definition of most types of individual social behaviour. Likewise, it is frequently seen that one behavioural pattern is designated and defined differently than another, uniquely because of the social context in which it is performed. In this way, "lateral spreading display" and "lateral parade" in certain cichlid fish include the same visual components, in terms of morphology and kinetics. All the same, the first is considered agonistic, as it is performed in front of a rival or in territorial conflicts, and the second, epigamic, as it is observed in the context of sexual synchronisation of a male and a female.

Each definition should be as clear and distinct as possible, so that interpretations and applications do not vary too much from observer to observer. Still, many parameters, often essential in order to define the units to be observed, are very difficult to evaluate in the absence of instruments for their measurement. For instance, distances between individuals, the speed of their movements or little-perceptible details which happen too quickly, or outside of the field of vision of the observer, can induce a degree of risk or arbitrariness. As in all operations which measure, the researcher must hope that any errors in measurement will be randomly distributed and not occur systematically in favour or disfavour of his hypotheses. Given the considerable number of different units to be noted, it is preferable to no retain behaviour which is of doubtful classification. The researcher must also make certain that the notation of certain units is not more difficult than others; if not, an unequal rejection of doubtful observation may introduce a systematic bias.

Figure 1. 

Once surface units have been defined, they will serve in the building of more in-depth units. In our research, certain concepts and behavioural super-categories were defined according to the superficial units presented in the Appendix A1.  As illustrated  in Fig.1, the definition of behavioural metaconcepts should ideally take on a tree-like structure, in which the leaves, or terminal units, are made up of behavioural patterns, the objects of direct observation, and the branches, of the supra and meta-concepts constructed from these patterns. So, an agonistic behaviour may include, by definition, aggressive or defensive behaviour, depending on whether the motor patterns of flight or submissive posturing are observed. It is the same for aggressive behaviour which, for example, in Xiphophorus, may include, first of all, menacing behaviour, itself divided into two observable patterns, namely, lateral beating and lateral display, and second, offensive behaviour, determined by observation of attacking and biting. We can see therefore that, starting with basic patterns which are directly observable, an entire series of more general and theoretical concepts may be built up, always remaining linked to the observable facts. Very often it is also necessary to define units involving interactions between individuals. These definitions may take into account the spatiotemporal context in which the behaviour is performed. For instance, an aggressive behaviour on the part of an individual may fulfill the definition of that behaviour, which consists in chasing another individual (which we may term a chase), if the latter individual responds with defensive behaviour (flight or submissive posturing). On the other hand, the same aggressive motor pattern may contribute to escalation if the initially aggressed individual responds in turn with aggressive behaviour. An escalation would first have been defined as aggressive behaviour on the part of one individual, concomitant or followed by aggressive behaviour on the part of second individual. These interactions (chasing, escalation) may in turn supply elements necessary to the definition of other, more comprehensive concepts. To continue with our example, the criterion for establishing hierarchical dominance between two individuals may be defined as being five chases (the concept being previously defined) successfully performed by on individual against a second, without this second individual successfully performing a chase in return. From these concepts, we may define others which are even more molar.

Repetitions and transitions

Some behaviour has a tendency to occur repeated manner or in bouts. Should we therefore note a single appearance of the behaviour, or as many different appearances of it? For example, suppose that a behavioural unit includes all the play activity performed by a child. Should the observer record the behaviour each time the subject manipulates the same object, or only when the child passes to a new one? Should the observer consider playing as if it were a state, having duration and being interrupted only by the performance of another type of behaviour belonging to another unit, such as demanding attention from an adult?

The problem of identification of the beginning and end of a type of behaviour, as well as its classification or its treatment as an independent event, is extremely important. the beginning and end of a type of behaviour may be specified in the definition of a behavioural pattern, which will indicate if a type of behaviour or a behavioural pattern will be considered as having duration or as being pulsated event. Also, if appropriate, will be indicated the criteria for determining the end of a behavioural pattern and the beginning of another within the same unit. The researcher may define two types of behavioural units depending on whether their duration is to be considered or not. Events are behavioural units of which the duration is of no interest: they are considered as having limited duration. For example, a bite, a peck, or a drinking action in a chicken are distinct types of behaviour, most often considered as events which are punctual limited, or without duration. On the other hand, for a researcher interested in the degree of stereotypy of a behavioural pattern such as the drinking pattern in chickens, each drinking action is considered as having a significant duration. This researcher would therefore do a detailed analysis of the movements involved, by, for instance, slowing them down on film. An event with a significant duration is sometimes but unproperly called a "state" (Altmann, 1974; Sackett, 1978), that is, the behaviour in which an organism is engaged. Durable events may always be, at the time of analysis, transformed into events without duration. Limited or punctual events, however, may not be attributed with duration afterwards, this information being irremediably lost. Separate analyses should therefore be undertaken, depending on whether durable or punctual events are being considered. Finally, a series of events noted as being a single appearance of a behavioural unit constitutes a bout of activity.

The determination of the end of a behavioural units, of a state, or of a bout of activity, is a delicate question. Used singly, or in combination, there are three kinds of criteria for determining interruption. First, performance of another type of behaviour may take place. For example, all the actions of a child manipulating an object will be noted as a single appearance of play activity as long as they are not interrupted by any aggressive or social behaviour. Second, an interruption may take place by means of a pause or an interval longer than that chosen as being necessary and sufficient for separating two sets of behaviour. Third, an activity may be interrupted by a change of object (such as picking up another toy) or place (such as grooming of a different body part).

Similar criteria may be applied to transitions in behaviour between individuals. Very often, the researcher is interested in determining the probability of one type of behaviour following another. the calculation of transitional or conditional probabilities is included in what is generally called analyses of behavioural sequences, each sequence being considered as a sentence from which the syntactical rules must be extricated. The problem is as follows: should we consider all the transitions between two types of behaviour performed by two individuals as belonging to a series of stimuli and responses, or should we impose a time limit beyond which a type of behaviour will not be considered as determined by preceding behaviour? The question may be decided by an ethological judgement, as shown by Dane and Van der Kloot (1964) in their study of behavioural sequences in the duck Common Goldeneye (Bucephala clangula). From their preliminary observations, the authors considered that any activity between two individuals separated by five seconds or less were part of a stimulus-response transition.

It is, on the other hand, possible to attentively examine results and decide in the light of more objective criteria. We may also examine the distribution of time intervals and of the logarithm of the survival of events belonging to the same behavioural unit in one individual, or of events belonging to two different units in two individuals (Machlis, 1977; Slater, 1974). In our study on the rules of aggressive exchanges in the Green swordtail fish (Xiphophorus helleri, Poeciliidae) (Beaugrand, 1997), an aggressive exchange between two duelling fish implied that one behavioural pattern followed another within a certain interval; for example, in the case of biting, the interval was 0.05 minutes. Otherwise, the transition was not recorded and a pause was obligatorily inserted between the two types of behaviour. This decision was taken after an examination of the distribution of intervals separating all types of behaviour, as well as the logarithmic curves of their survival. Ideally, the chosen criteria should reflect the organization or the structure of the behaviour itself (Slater, 1974), and an analysis of these curves may uncover the underlying processes. An histogram of the intervals separating events which occur randomly conforms to a negative exponential distribution, following a Poisson distribution (Cox and Lewis, 1966), and is characteristic of distributions in which the measures are temporally independent, Fig.2 presents such an histogram of observed intervals between bites, performed by two duelling fish during planned combats. It is evident that the very large majority of intervals does not exceed 0.04 minutes. It is difficult, however, to decide upon a criterion for determining interruption from this data alone.

Figure 2.  A logarithmic curve of survival may be obtained by establishing, for each interval, the total number of observed intervals which are of still greater duration. The natural logarithm of this number will then replace, in the histogram, the observed frequency of the corresponding interval. To illustrate: on 133 occasions, the interval measured between bites on the part of two individuals exceeded 0.04 minutes, and the logarithmic of 133 is 4.8903. Fagen and Young (1978) cover in detail the application of this procedure. In principal, if all measures are independent, and randomly distributed, the logarithm of the frequency of perpetuation should take on the configuration of a straight line. As seen in Fig. 2c, all straight lines drawn between two points indicate a slope directly proportional to the probability of a bite occurring after a given interval. Any changes in the underlying probabilities should manifest themselves by a sudden change in the shape of the curve, which would suggest the existence of a different process before and after the change. So, we may see in the transitions between bites performed by two individuals, a reduction of the slope between 0.04 and 0.06 minutes, and therefore a substantial decrease in probability. This would incite us to consider as transitions from one bite to another only those associated with intervals not exceeding 0.05 minutes, and to consider any others as necessarily interrupted by a pause, and not dependent upon the same process. The same transformation may be applied to the transitions between bites of only one individual (see Fig. 2d-f).

Update: Slater & Lester, 1982 suggest to take the found interval plus 5% to be sure to include all relevant data.

    There is a more mathematical technique for establishing if the researcher is dealing with only one distribution or many partially overlapping ones. It consists in comparing the observed distribution to a series of theoretical gamma-type distributions (Derman et al., 1973) and finding an optimal separation point between two or more amongst those which are most conforming to the observations. This approach has been utilised by Lefèbvre (1982). Other objective techniques have been described by Law and Kelton (1982).

    When analysing transitions in sequences, it is unusual to give priority to the time intervals between one element of the sequence and the next. It seems essential, still, that the transitions constructed should respect the natural distribution of events. For each type of behavioural transition observed in a single individual or in two individuals, the logarithmic curves of the survival of the time intervals should be studied in order to establish a temporal criterion beyond which two behaviours will be considered as independent, statistically and, we presume, ethologically. the analyses of transitions which respect such criteria will furnish a clearer idea of the processes involved.

 

THE DESIGN OF OBSERVATION

    The recording of behaviour by means of direct observation should be rigorously planned. With the goal of assuring reliability, the plan of observation involves the methods by which the measurements and observations will be carried out, as well as the methods for controlling certain variables. The elaboration of the plan takes into account the following three types of considerations: the research objectives, as well as the presence or nature of hypotheses to be tested; the possibilities of carrying out observations, depending on the number of behavioural units decided upon, and the techniques available which may be of use; and finally, the precision, capacity, and efficiency aimed at. In finalising his or her plan of observation, the researcher takes a series of decisions concerning the eight points we shall now examine.

 

The choice of taxonomy

Which behaviour in the subject's repertoire is to be observed? The answer to this question is most often dictated by the researcher's objectives; are chosen only behaviour pertinent to these objectives. Lacking precise questions or hypotheses, however, it is difficult to judge what is or is not pertinent. Which types of behaviour should be recognised within the whole of the observable behaviour, and which types should be noted or encoded? The distinction between that which is recognised and that which is recorded is not an artificial one. The behavioural patterns of a species constitute a superficial taxonomy, from which the researcher may construct a more molar taxonomy, a deeper, more underlying one. In practice, it is very unusual, when a researcher is seeking answers to precise questions, that the behaviour is noted in terms of the units constituting the superficial taxonomy of the species, unless, of course, the questions specifically regard this taxonomy. More often, the notation will be more molar. In this way, only an aggressive behaviour will be noted, instead of recording the specific behavioural unit included in this super-category (for example, a bite or slap); we may also take note only of the asymmetric result of, for instance, a quarrel between two adolescents, one having the upper hand.

The extent to which data is encoded depends equally upon the safety margin allowed by the researcher so that he may afterwards reconsider the categorisation. If the encoding is highly detailed, the risk of error will increase, and the research will lack efficiency. It will also encourage a compulsive data "sniffing" in the absence of any hypotheses. On the other hand, it can be interesting to partially alter the classifications to extract hypothetical regularities which may be tested afterwards. If the encoding is very molar however, there will be a danger of losing a certain sensitivity in the information. This sensitivity is measured in relation to the hypotheses of the research however, and cannot be discerned by any absolute means. The researcher should also decide if the behavioural units chosen shall be temporally limited events or events with a significant duration.

Recording of temporal elements

    Should the researcher, as well as noting the identity of each behavioural pattern, record the time at which it occurs? Should he or she limit himself to recording behaviour in the order of its appearance, or even to enumerating behaviour so as to establish its frequency? Here again, everything depends upon the questions being asked and the practical and technological considerations governing the research. The decision goes hand-in-hand with the choice of sampling techniques and coding devices to be used. It is evident that the conclusions which may be reached will be different depending on whether the underlying analyses concern frequency or duration. The continuum upon which the duration of behaviour varies extends beyond the continuum related to the frequency of behaviour: we may therefore expect duration to be more sensitive to differences than frequency, although it may be more difficult to interpret. It is therefore appropriate, whenever possible, to record both the frequency and the duration of behaviour. This is made possible thanks to microprocessors designed to record coded observations; we will call these instruments ethographs. The majority of ethographs on the market automatically record temporal elements, and this, with sequential recording done during the continuous sampling of behaviour, permits numerous supplementary measurements which could not be rendered possible by using recording techniques which are limited to noting the appearance of behaviour, or even the sequence of its appearance. Some of these supplementary measures are presented in Table 1.

Table 1. Comparison between the various ways of noting behaviour as regards to the measures they can furnish. The following information can be noted: Who: which individual initiated the behaviour; What: which behaviour unit it used; Order: its order in relation to preceding ones; Duration: its duration; Moment(t): time of occurrence. When there are more than two individuals interacting, the target individual can also be noted (when only two individuals interact, the target is known implicitly when the initiator is known).

Methods of recording

The quality and representativeness of observations also depends upon the method of recording. Should observation be performed live, while the action is taking place, or should it be deferred, taken from video-tapes, films or tape-recordings? The answer depends, again, on the research objectives. In certain cases, it may be necessary to store material for future decoding: this, for example, is the case when analysis must proceed by means of slow-motion filming. The use of video-tape or film obviously makes possible the repeated review of observations according to a variety of observational criteria, and permits a very detailed and fine analysis of behaviour. The technique is not without its drawbacks however. In the first place, these methods of recording only provide a two-dimensional and sometimes monochromatic representation. They also most often require special lighting, and higher expenditures. Unless wide-angle lenses are used, some information will be lost. Moreover, the presence of all the necessary equipment cannot help but perturb the subjects, especially if they are human. The last, but not least of the dangers is that of a compulsive data sniffing without reference to pertinent questions.

Encoding and Notation Systems

It is not always necessary to be encumbered with complex instruments in order to make very valuable observations. Paper and pencil are often sufficient, especially if the research objectives are clear, and if that which is to be noted is sufficiently molar. Checklists, matrices, meters, recorders or even polygraphs may be used. Using instruments directly compatible with a computer greatly increases research possibilities.

In any case, a code system will be used. Made up of alphanumeric or pictorial signs and symbols designating pertinent units, it increases speed and efficiency in recording observations, in validating them, in transferring them to a computer, as well as in analysing them. Well chosen, the system may be treated directly by computer. Likewise, the shorter is the code, the more rapid is the recording of it, and the smaller is the space it occupies in a record book or on a computer disc. If the code is visually, auditorily or mnemonically evocative of the behaviour it signifies, the observer will have less hesitation in encoding this behaviour and less errors will be committed. Encoding aims at facilitating observation, not hampering it. It is obviously dependent upon the explicit rules and conventions which support the classification of observations. Each pertinent classification or unit to be recorded should have a specific corresponding code which is unique and exclusive. Encoding goes hand-in-hand with the sampling techniques chosen, and with the dimensions of what is to be recorded (duration, beginning, end, intensity, origin, destination, orientation). Table 2 presents the code system which we have applied to sampling by successive focalisation upon many members of a group. If the duration of certain types of behaviour is required, special codes may be used as prefixes or suffixes of behavioural codes, one indicating the beginning of a type of behaviour, the other its cessation. If such a system is properly used, we may take note of concurrent or simultaneous behaviour. Golani (l976) has developed a specialised vocabulary with which to describe concurrent behaviour. We may find a good summary of this in the Handbook of Ethological methods by Lehner (1996).




Table 2. Example of a simple encoding system using a static recorder.

    The use of a sophisticated system of recording such as The Observer greatly facilitates the task of defining codes, validating and editing entries. It is a professional system for collection, analysis, management and presentation of observational data using a desktop or handheld computer. It can even edit and analyse behavioural processes from video.

 

Complete and continuous observation as opposed to sampling

    Short of completely recording all behaviour, the researcher should, as well as choosing the behaviour to be recorded, decide upon the time periods during which these observations will be performed. The reliability of the observations, their usefulness and their capacity to lead to generalizable results, all depend upon temporal sampling. Sampling may be done on a regular or irregular basis. It may be initiated by events which particularly interest the researcher, or even by circumstances (such as a change in environment or state) which he or she would like to put into contrast.

 

Backup techniques

    The researcher should make an appropriate choice of techniques and tools which will help in the encoding and recording of observations. There is a wide range of variety in these techniques, and the choice depends on the questions posed by the research, and the computer skills of the researcher or his or her helpers. The method of pen and paper, demonstrated by Lehner (l979) in a number of examples, remains the simplest and least costly. On the other hand, the range of research questions which may be approached by this method is very limited. Electric meters and chronometers, video or tape recorders may be used, as well as polygraphs; but if their data is to be computer processed, the observations must first be re-encoded into suitable form for processing. This re-encoding implies an enormous amount of work (viewing, transcribing, feeding the data to the computer and verifying that data). It is easier, if one wishes a detailed analysis, to use ethographs.

    For some years now, a number of electronic systems have been developed which aid in the gathering of observations in field and laboratory. The most well-known are the Datamyte (900 and 1 000) ad hoc systems by Electrogeneral company, and the OS?2 and OS?3 by Observational Systems. But Radio-Shack portable C?100 can easily be programmed to do the same task. These instruments permit the encoding and recording of the appearances of behavioural units, as well as real-time duration, and the production of lists which may be either directly transmitted to a computer, or temporarily stored on cassette or disc for future use. Thanks to these instruments, the observer can encode behaviour live, or indirectly from video or tape recordings. These microprocessors are battery powered, permitting a functional autonomy of up to 24 hours if there are pauses between sampling periods.

In deciding to use ethographs, the researcher should be aware of the consequences of his choice, which will necessarily incur intensive use of computer resources. The researcher must either be equipped with the required competency, or assured of the services of a programmer. Moreover, use of ethographs calls for considerable financial investment, not only in the purchase of hardware, but in the training of observers and in the development of a reliable code of observation.

    On the other hand, the positive spin-offs are important. The first, without question, is the ease with which observations may be performed. The range of information obtained is greatly enriched, particularly with regard to the temporal and sequential organization of behaviour. All the same, to extract this information, computer processing is necessary, for the transmission of observations to larger computers, and for the validation and refinement of these observations, as well as for the management of the files containing them and their analyses. Direct observation of behaviour produces enormous quantities of data which must be stored economically. The files containing intermediary results and the summaries of them must also be stored and archived.

A second advantage lies in the speed with which analyses and summaries may be produced. Once the means of data-transfer between the ethograph and the central computer have been established, that observational errors have become scarce, and that validation programs are in operation, observations gathered in a day may be analysed and summarised within minutes after data-transfer. This, of course, rests on the supposition of an efficient human and technical substructure, particularly for software preparation and maintenance, and of corresponding financial means.

    The third advantage is of a quantitative and statistical order. Data may be easily analysed with the help of currently used software. Its use, however, entails a high degree of statistical competence on the part of the researcher, or constant access to an expert in the field. It also requires a healthy pragmatism in regard to the identification of problems more economically resolved by humans than through software. It calls for the capacity to be satisfied with the elaboration of small programs, each accomplishing a precise but limited task, rather than a vast orchestration of imposing software, which will never work correctly without drawing on the entire memory of the machine. Finally, the researcher must be immunised against the tendency of compulsive sniffing data, and that of analysing data without reference to precise research questions. If computer processing is widely used in research, its logical and technical contribution should not lead to underestimating the weight of the researcher's intellectual planning.

 

The degree of generalisation

When employing direct observation of behaviour, even when experimental methodology is being used, the researcher is traditionally more concerned with the external and ecological reliability of his results than when approaching the behaviour in question by another means. It is important for the researcher to decide, before undertaking his study, upon the degree of generalisation which is aimed at. If it is a high degree, a great number of situations and different time periods within an individual's life must be sampled, as well as a great number of individuals. The danger in this case is that the researcher may be unable to easily extract general regularities, as the observations may be highly variable. By making the observational situation uniform and consistent, the variability inherent in observations will have a tendency to decrease, and regularities will emerge more clearly. It goes without saying, however, that this advantage brings with it a lesser capacity to generalise the results for situations which have not been observed. The researcher should also arm himself or herself against an unequal observation of subjects, some of which may be observed for longer periods, at unrepresentative moments, or again at moments which favours or disfavours them in relation to other subjects. Likewise, some periods in the lives of individuals may be exaggeratedly observed, to the point of creating results which are no longer representative of the usual behaviour of the species.

 

Factors to be controlled

The goal of a sampling technique is, above all, to provide measures which are precise, valuable, and representative of behavioural regularities. The technique foreseen in the observational plan should assure a certain representativeness in the observations made. Notwithstanding, its rigorous application does not, alone, guarantee the validity of the research. External controls should be set up, even for field work. They will most often be provided by selecting the time period during which observations are to be made (Schneirla, 1950). All factors liable to contaminate or invalidate the representativeness of the observations should be neutralised, insofar as the control operations will not influence the events observed. Certain contaminating factors should be particularly watched for: effects due to observer interference, expectations on the part of the subjects or observers, and instrumental inconsistencies.

Effects due to observer interference

The observer must assure himself, herself that he, she in no way influences the events happening before him or her, for example, by his, her presence, to which the subjects may certainly become gradually accustomed. If he, she does exert and influence, it should be exerted equally under all observational conditions, to avoid any artificially produced variations. In some cases, the observer should be concealed. On the other hand, a hidden observer might appear more predatory than if he, she were to be unconcealed and allowed the subjects to get used to his presence. Also, captive and semi captive animals may perceive the observer as a generous food dispenser. Long-distance observation using cameras, telescopes and binoculars, or observation from behind a one-way mirror may be of great help. When working with human subjects, one must, however, receive their consent (or the consent of those responsible for them), after giving frank and sufficient information as to the nature of the research. This is a basic deontic rule which must be respected even if it conflicts with the principles of non-interference already discussed.

 

Expectations of subjects or observers

When using humans as subjects, the research situation is never free from a desire to support the researcher's hypotheses, if they are known, or to satisfy the expectations of the observer, as they are perceived (Christensen, 1977, 1979: see Christensen, 1980).

Likewise, observers are not exempt from expectations which lead to not recording a behaviour which did, effectively, take place, and recording another which did not. Although unintentional, the bias is nevertheless very real. It seems to enter, most often, at the time of observation, and during the reclassification of observations. The errors involved principally concern behaviour which is doubtful or obtained under unorthodox conditions. Rosenthal (1978) has estimated that in a research project, the observer is mistaken in classifying approximately 1% of his observations, and that 66% of these errors serve to support the research hypotheses in question. Although this phenomena is evidently not sufficient to significantly reorient the conclusions of study, it is the researcher's duty to reduce its presence. Researchers who are aware of this possibility or error at the time of recording and reclassifying, and who make efforts to eliminate it, succeed to a great degree (Rosenthal, 1978).

Another source of error attributable to the observer may occur when the researcher interprets his, her observations, that is, at the time of classification of behaviour, particularly if the criteria for classification are abstract and therefore causal or functional. If the classification is established during encoding, the interpretative errors are irreparable. On the other hand, if the classifications are established a posteriori from an initial concrete descriptive classification (according to form or effect), the errors in a second classification may always be eliminated by determining another classification with the aid of a new theory. Observations made according to concrete descriptive criteria possess a factual character which is assuredly less disputable than that of observations made according to abstract criteria.

It is possible to avoid these errors, and this bias. In the first place, the researcher should give much weight to the importance of establishing the list of observational units. As already mentioned, it is preferable to use concrete and descriptive criteria for classification, and to reserve classification according to abstract and theoretical criteria until the transformation of observations into data comparable with the empirical implications of the hypotheses. Second, the basic units must be defined so that each contains a clear and exclusive criterion. In case of doubt, an event should not be retained, and should not be forcibly included in a particular classification. In the third place, it may be possible to employ well trained observers who are ignorant of the researcher's hypotheses. If experimental manipulations are performed, they should be performed by someone other than the observer.

Instrumental decay

It is well known that instruments may perform in a variable manner, sometimes working better, sometimes worse. In the case of direct observation, the instrument consists of an observer who directly interprets a list of behaviour which has been judged pertinent. It is possible that this behaviour may not have been clearly defined, and that the observations lack precision. It is also probable that the acuity of the criteria for recognising and accepting behavioural patterns may dwindle in the course of observation, because of fatigue or lack of attention, or, contrarily, augment from one period of observation to another, because of increased skill on the part of the observer: consequently, the observation will lack stability. This inconsistency in an observer should be controlled, likewise for inconsistencies between a number of observers. As we shall presently discuss, a minimum of reliability in the work of the observers should be assured.

 

Example

The application of various methods of control in the direct observation of behaviour may be most clearly illustrated by an example. Each square of the checkerboard presented in Table 3 corresponds to one hour of continuous observation, during which the observer records all pertinent behaviour performed and received by an individual upon which he, she temporarily focalises his, her attention. Each of four individuals within a family is observed in turn. We are interested in the aggressive interactions which occur between the individuals, and in the hierarchical organization within the family. During one hour of observation focalised upon one individual, all the behaviour which he, she performs, the individuals in regard to which he, she performs them, all behaviour directed towards the observed individual, and the identity of the individual who initiates the behaviour are noted. Using a systematic sampling technique, a number of behavioural variables are simultaneously observed, each with a number of dimensions (intensity, frequency, duration, direction and origin). The danger of compulsive sniffing of data is consequently multiplied proportionally. We may see that the order in which individuals take their turn as the focus of observation varies randomly from day to day. In this way, there is an assurance that differences between subjects are not introduced by a regularity in the order in which they are studied.

 Figure 3. Plan of observations to be completed out over four consecutive days. The plan equalises subjects and families (or treatments) over the number of periods of observation (and total time observed), and randomises the order in which they are to be observed. Observers are also randomly affected to observation sessions, which are made of constant duration. Black squares correspond to observation sessions.

The plan of observation also foresees the number of times each group of four must be observed. In the example of Fig. 3, the observations are repeated over four consecutive days, but they could have been done according to a different method. If the plan foresees many observers, it should also specify how these observers will be associated to the different groups and the different periods of observation. In this way, each family might be observed by a single observer. Over the whole of the research, however, the observation of different families might be randomly distributed to the various observers. The plan of observation indicates who to observe (individual or group), when to observe, the duration of each sample and the total duration of the observations, and the order (arbitrary or random) in which subjects or groups are to be observed. Also, the plan mentions the number of times a series of observations will be made upon a single subject or a single group. Moreover, it foresees how and when the observations will begin and end. It provides the behavioural units which will serve as an exploratory model for the research. Finally, it indicates how intra-observer and inter-observer reliability will be assured and measured.

SAMPLING TECHNIQUES

Altmann (1974) discusses in a detailed and critical manner each of the sampling techniques used in direct observation of behaviour. Lehner (1979), Sackett (1978), as well as Hutt and Hutt (1977), also present good expositions. The techniques described here are not presented according to a classification within a single parameter (for instance, continuous recording as compared to discontinuous, or random recording as compared to that focused upon a specific individual); the classification used is that which corresponds most to current terminology in the field.

 

Non-structured or ad libitum sampling

Non-structured or ad libitum sampling is used by Sunday observers and amateurs, those who love the animals their subjects (Lorenz, 1981). No restraints are imposed in regard to the subjects, the order of observations, or the periods of observation. It is in fact inappropriate to speak, in the context of this chapter, of a sampling technique at all. Only the most spectacular, the easiest, the most interesting events are observed. Producing notes and sketches jotted down during the action, this method of observation is a preliminary to any systematic observation, or to any descriptive ethological work. this method is implemental in discovering natural behavioural patterns leading to principal questions and hypotheses which may eventually be more rigorously studied. Observations made using this technique can never be subjected to a quantitative preliminary analysis.

Complete and continuous sampling

Complete and continuous sampling produces the most detailed recording, from which other, more fragmentary records may subsequently be drawn. In this type of sampling, the observer notes the type of behaviour, the identity of the actor, the time of occurrence and the duration of all behavioural units which are of interest. In some cases, it is possible to record all behaviour, within a restricted repertoire, performed by a limited number of individuals. It is not necessary to record everything live; it is quite possible to view and re-view, a number of times, a taped recording, and to extract, each time, a part of the total behaviour. Also, two or more observers may simultaneously observe the same interactions, each person recording the behaviour performed by only one individual. In principal, complete and continuous recording may be reconstructed by synchronising the various vectorial sequences formed by the observations in order to determine one single sequence, made up of all events recorded. In practice, the procedure becomes complicated because of the simultaneous and overlapping nature of certain events. In some cases, it is possible to do live continuous and complete recordings, if a small number of behavioural units and individuals are to be observed, and if the behaviour to be studied does not occur too quickly. The recording thus obtained is identical to that obtained by the technique of successive focalisation, to be presented shortly; the initiator and the receiver are always the same two individuals (for instance, a father and his child, or a couple of wolves). Complete and continuous recording is the richest source of information, especially if the duration of the samples is sufficiently long, and if the times at which the behaviour occurs is noted. This enables measurement of such elements as the rate or frequency of certain behaviour and changes in these rates, the frequency of transitions between sequences, the duration and latency periods of the first behaviour to occur, the breakdown of different activities (information classifying the relative importance, in terms of frequency or of duration, of each behaviour in the repertory), as well as many other elements derived from those mentioned. This form of recording necessitates the use of an instrument (polygraph or ethograph) which permits the automatic recording of the temporal elements present in observations. Checklists or tape-recorded dictation are too limited and imprecise to serve the purpose.

An example may illustrate the nature of the technique and the wealth of information which continuous sampling makes available. In a recent work, the author has compared the agonistic behaviour of fish which were known to each other, having met in duels, to that of fish unknown to each other, but having profited from recent and comparable social experience. We predicted that, when put again in the same aquarium, individuals in the first group of fish would recognise their former opponents and immediately re-establish, without agonistic exchange, their former dominance relations. It was foreseen, on the other hand, that the other fish would fight in order to establish a dominance relation, which would entail numerous aggressive exchanges. Table 3 presents two segments of recording representing the encounters of fish which previously knew each other or not, as well as various measures extracted from a number of similar recordings. The observations were carried out by only one observer who noted all interactions with the help of an ethograph, which automatically recorded the time throughout the observations. The recording started as soon as the fish were placed in the same aquarium, and finished when one individual could be recognised as dominant over the other, according to a specific aggressive dominance criterion. As shown in Table 3c, very clear differences were indicated, attributable to treatment, in the frequency of the appearance of aggressive behaviour.

Table 3. The syntax is "initiator fish, act by initiator, target fish, act by target, time since beginning of observation session". Thus "1 APP 2 IGN 387 " means that fish 1 approached fish 2 which ignored (did not react to the approach); this occurred 387 hundredth of seconds after they were introduced to each other and the observation session began. Abbreviations:  APP: approaches; IGN:  ignores; MEN: menaces (lateral displays, tailbeat); UP: rises from bottom; IMM: the fish drops immobile on bottom; DEF: adopts a defensive posture or flees; OFF: the fish attacks or bites; FLU: the fish flutters alongside the aquarium sides. The dominance criterion was set to five unilateral approaches, menaces, or attacks initiated by a given fish, systematically followed (or concomitant with) by defensive behaviours by the alternate fish.

Sampling by successive focalisation

Sampling by successive focalisation supplies a continuous recording, for a limited time period, during which the observer focuses on one individual or on a group. The observer takes note of the time at which behavioural units appear, or simply the order of appearances, or the transitions between them. This involves observing, during a predetermined time-span, all pertinent behaviour performed or received by the object of the focalisation. All individuals, or only certain members of a group, are successively observed in the same manner, and for the same amount of time. Sampling or non-social behaviour is a relatively simple matter. For example, we can, during ten consecutive minutes, take note of the number of times an individual animal feeds itself. The sampling of social behaviour patterns is, however, much more complicated: for each behaviour noted we must also indicate who performs it and to whom it is addressed. An example will allow us to illustrate the procedure in question. In one of our researches (Beaugrand et al.,1984), we wished to test a hypothesis concerning the existence of a despotic social organization and of a harem, in fish. Each member of a group, consisting of four males and four females, was successively observed by focalisation for 15 minutes a day over four days. The method of encoding allowed for identification of the initiator of a behaviour, and the type of behaviour performed by him, and, when applicable, the identity of the receiver and the behaviour emitted by him. The time lapsed from the beginning of the focalisation was also automatically recorded. Figure 4 presents a summary of the behaviour and the interactions noted during the two hours necessary to successively observe eight individuals; we may see that the despot (individual 4) affirmed his aggressive dominance over all the other males (individual 1, 2 and 3). The despot also courted with all the females, with the exception of female 7, which was exclusively courted by individual 2, of second rank in the hierarchy. It must me remarked that the information obtained concerning the behaviour of a given individual does not come uniquely from the time period in which this individual served as the focal point, but also from all the interactions performed with another individual, when the latter was being focused upon. 

Figure 4. Sampling by focalisation results.

Sampling by focalisation is a particularly useful technique in field-work, especially if the identity of the individuals observed is not critical, and if all individuals cannot be observed simultaneously, if only because they are not all visible. In choosing, randomly, which individuals will be observed, the researcher may arrive at adequate results, concerning, for instance, subjects which differ in age, sex, or socio-economic level. Subjects may also be chosen according to an order imposed a priori, or by a research plan. In studies in which it is essential to establish the identity of individuals, the disappearance of the observed individual from the observer's field of vision poses a problem for which no satisfactory solution has been suggested. If the disappearances are brief, we can always lengthen the period of focalisation on the individual by an equivalent amount of time, but if they are long, the observations gathered must, most often, be disregarded.

The technique of sampling by focalisation, when accompanied by a pertinent choice of behavioural patterns to be observed, and sufficiently long focalisation periods, is as efficient as complete and continuous observation. In most cases, it is also the most economical method. According to Altmann (1974), it is the most efficient and profitable technique.

Sampling in Sequence

Sampling in sequence is used when the researcher is principally interested in the order of different types of behaviour within a relatively regular chain. This is the case in parades, greeting ceremonies and agonistic and epigamic rituals. The sampling period may begin at the initiation of a sequence and end with it. All types of behaviour that it is possible to identify as being part of the sequence are systematically noted in the order in which they occur, until the sequence is finished or interrupted. A new series of observations is commenced when a new sequence appears. This technique, therefore, involves data driven, i.e. recordings of which the beginnings and ends are determined by the facts themselves, and not by some external signal marking the end of a certain time interval. Most researchers using this technique do not record the time of the initiation of the sequence, but only the order of events within the chain. It is, however, possible to record the time at which the behavioural units occur if the units are performed very slowly, or if the events may be recorded on film, to be studied in slow motion.

This technique has very limited applications. It is rarely used in live recording. Hazlett and Bossert (1965) made use of it to describe agonistic sequences in hermit crabs, and to determine the degree of association between the elements of the sequence; still, as in the majority of cases, the synchronisation of the beginning of the sample with the beginning of the behavioural sequence was established a posteriori from continuous or focalised recordings, at the time of analysis. We may also find in the work of Lemon and Chatfield (1971) an analysis of the cardinal's song, constructed from sequences extracted, in this way, after-the-fact. The principal difficulty apparent in this technique is in the choice of criteria delineating the beginning and end of a behavioural sequence.

Sampling by presence or absence

Sampling by presence consists in noting if a behavioural units is present or absent during a short period of time. It is used, above all, to record states, which entails behaviour with significant duration. Many different behavioural units may be recorded during one time-period. For each period, the observer notes if the observed individual performs, at least once, a certain behaviour, or is in a recognisable state, such as eating or working. The behaviour of one or many individuals can thus be noted as being present or absent during each period of sampling. These periods are usually very short (15 to 30 seconds), but numerous and performed at regular intervals, for instance upon emission of an audible signal.

This technique does not provide measures for duration or frequency of behaviour. Altmann (1974) advises against its use. In some cases, we may nevertheless see in it a certain utility for determining the relative importance of different types of behaviour (frequency and duration). Information relating to each behavioural unit or each state, in other words, the number of times that this unit was present upon initiation of sampling, is probably a more realistic indication of the relative importance derived from total frequencies or total duration of a behaviour (Baerends et al., 1970; Ollason and Slater, 1973). It is still preferable (Adams and Markley, 1978) to draw this information a posteriori, from continuous and complete recordings, or from those obtained through the following technique.

Sampling by instantaneous scanning

Observation by instantaneous scanning proceeds in essentially the same manner as does sampling by presence or absence, except that the sampling period is instantaneous, and theoretically in a given moment. If many individuals are to be observed, each successively becomes the object of a visual scan: the observer notes in which state or activity each individual is momentarily engaged. It may be preferable in certain cases to observe spaces rather than individuals. To illustrate: in our research concerning the existence of a despotic system with harem in green swordtail fish, the spatial positions of the eight individuals in each group were recorded every 15 minutes, over two hours. We had foreseen that, on the whole, the despots would be more frequently found in proximity to the females than to the other males in the group, and farther away from the subordinate males than the subordinate males from each other. The aquariums had been gridded, and at the end of each 15 minute period the instantaneous co-ordinates of each individual were recorded. It was seen to be more practicable to observe each sector of the aquarium to establish the presence of individuals, than to establish the co-ordinates of individuals after locating and identifying them. Table 5 presents the co-ordinates obtained during eight visual scans of the group, comprised of the 4 male and 4 female adults studied in Fig. 4. From these eight scans, as many distance matrixes between individuals were extracted, which were then summarised in a single matrix, using multidimensional spacing (Kruskal, 1977), and translated into a three dimensional space, illustrated in Fig. 5. We can see from this figure that the spatial disposition of individuals, made evident when placed in a single dimension, is essentially the same as that obtained by calculating, for the entire observational period, the total time spent by the individuals in each section of the aquarium.

If Altmann (1974) identifies other dangers in using this technique, the main problem in instantaneous sampling, paradoxically, is that it its not sufficiently instantaneous. In practice, the more an observer lingers over an individual, the more the others are likely to change their state, posture, or position, and the more the product of the observations will resemble that of a series of samples, of inconsistent duration, successively focalised on various individuals. To minimise this drawback, a photographic technique may be used. This type of sampling is, nevertheless, very useful to determine the percentage of time which each individual devotes to certain activities.

Table 5. Successive scans taken obtained of the instantaneous positions of four males and four females in an aquarium divided into a large and a small compartment illustrated in Fig. 5. Data represent the coordinates in the aquarium at which the corresponding fish stood when the scan was made. For instance, male 1m stood at coordinates 321 at all the 8 scans, i.e. in the small right section of the aquarium, in depth in middle up zone, and in the back of the aquarium. From these coordinates, distances between individuals can be calculated, and synthesised using for instance multidimensional scaling.

Figure 5. (a) Geometrical representation of the spatial association of 8 types of fish. The tridimensional representation of the individual fish was obtained by the application of multidimentional scaling (KYST) to inter-individual distances obtained through repeated instantaneous sampling periods in which both female and male hierarchies were complete. (b) Total time spent in the large section of the aquarium by the corresponding fish individuals. The results obtained by these different approaches coincide, though one is based on instantaneous scan sampling and the other on continuous measurement (Based on Beaugrand et al., 1984, Behaviour, 91, 24-60).

Matrix completion

Strictly speaking, matrix completion is more a manner of determining and placing asymmetric relations in tabular form than it is a technique of sampling. It may be applied when relations between individuals are being recorded and not individual behavioural units. The asymmetric relations in question may concern, for example, dominance, spatial proximity, communication, affiliation or attention between the individuals of a group.

Matrix completion usually begins with the entry of the most evident relations, those easiest to record. Missing relations are then systematically sought, in order to complete the matrix. The following example illustrates the nature of the procedure. A researcher wishes to determine the order of aggressive dominance relations existing in a group of children in a day-care centre. He /She has defined, beforehand, a certain number of agonistic interactions which, when resolved in favour of one child, indicate that that individual dominates the losing individual. A criterion for aggressive dominance will have also been defined: for example, five agonistic interactions settled in favour of one child suffice for him, her to be declared dominant over another child. A matrix keeping track of the interactions of all the children is presented in Table 6. It consists of as many lines and columns as there are individuals in the group. The lines and columns represent, respectively, the individuals declared winners and losers in isolated interactions. Each time an interaction between two children is clearly settled in favour of one, it is recorded in the space at the intersection of the line and column corresponding to the subjects concerned. 

Table 6. Dominance matrices. (a) A first matrix is completed by keeping tracks of dyadic aggressive interactions in which a child expressed superiority over another during continuous observation periods of 4 children interacting. The criterion of dominance of one child 

over another was set to 5 unilateral and successive "superiorities". The moment of fifth occurrence was noted. For instance Martin, retreated or conceded objects to J. Fred on five successive occasions, the fifth one at 14h10. (b) From (a) a dominance matrix is obtained by substitution of all declared dyadic dominance relationships by a 1 in the cell corresponding to each pair of individuals and by rearranging rows and columns such that all (or most) 1s appear above the diagonal, and row totals correspond to the underlying presumed hierarchy. 

Most often, completion of such a matrix aims at revealing a comprehensive structure translating the relations between all individuals, and not only between those who are more active than others. It is the case in the example just given: complete hierarchies are being sought. We would attempt, therefore, to record as many interactions as possible without the restrictions of an order of observation or a uniform time-period being imposed by the plan of observation. In our example, as soon as the dominance criterion has been satisfied, the observer will direct all her attention toward the relations not yet determined, and will focus on the individuals concerned, who are often less active.  Sometimes the researcher, when filling in the matrix, notes all the interactions observed during a given observational period. In such a case, however, he or she would not be able to compare the frequencies obtained in one space with those of another, nor to consider the matrix as a contingency table.

Choice of a technique

Many of the techniques presented her may be combined to create a composite technique. They may also be used together, respectively, in two or many observational periods. Is it necessary to call to mind once more the importance of the questions asked by the researcher and the hypotheses formulated by the research? They determine the choice of sampling techniques, and these techniques themselves help to determine the validity of a research project. IN the absence of any question or hypothesis, only one technique can be used, that of non-structured sampling ad libitum. This allows the researcher, on the one hand, to acquaint him or herself with the species studied and the conditions under which observation may take place, and on the other hand, to identify the regularities inherent in the species' behaviour, which may lead to questions and variables, likely to be the objects of a systematic and quantitative research.

Table 7 summarises the cases in which each of the sampling techniques is recommended. The use of each technique produces different results, which are often not comparable with each other. In a comparison of seven different techniques, used to answer the same questions, Dunbar (1976) has shown that the results determined by each technique were only comparable on the ordinal level, which is to say, only with regard to the relative importance of each of the various measures taken using a given technique. Still, it is rare for a researcher to be interested only in the ordinal relations between the individuals of a group, or between different types of behaviour. Dunbar's work clearly illustrates the importance of the choice of a sampling technique. It is a theoretical choice, made on the basis of the definition of the measures sought by the researcher, these definitions themselves being suggested by the research hypotheses. It is also a practical and subjective choice. Before undertaking a research project, it is prudent to verify the validity of a number of techniques in answering the same questions, and the possibilities of applying these techniques.

The information received from observation will also differ according to the duration of the observational periods and the number of times they are repeated. For instance, if the order of the transitions in a behavioural sequence is sought, it is preferable to greatly lengthen the observational periods and to observe different individuals a number of times. 

Instrumental accuracy

Part of the validity of a research project employing direct observation of a behaviour depends on the accuracy of the instruments used, that is to say, on the reliability of the observational data produced. The observer acts as a filter, detecting the presence of such and such a pattern of stimulation. The reliability of the observational data he or she produces depends upon the observer's consistency, precision, and sensitivity, and is the result of an interaction between the observer and the list of behavioural units which have been judged pertinent to record. The observer interprets a series of definitions of these units, each definition corresponding to an exploratory model. As we have seen, the observer may bias his or her observations in a number of ways. The complexity of the system of observation, the number of units to be recorded and the number of dimensions of each unit (time, origin, direction, destination or intensity) may greatly influence the reliability of observations.

Information resulting from the use of an instrument may be consistent and accurate. It will be consistent or stable if it is identical from one application to another, the observer consistently interpreting the list of definitions and making identical observations of identical facts. It will be accurate or precise if there is appropriateness or correspondence between what is reported and what in fact occurred. In a psychometric or ethometric sense, instrumental reliability designates, according to Kerlinger (1973), the degree of error in measurement produced by the instrument used to measure. If two different observers are required, they should form one single, homogenous instrument. Consequently, instrumental accuracy may easily be doubtful, when in fact this accuracy is the basis of the validity of all measures taken. Therefore, every effort must be made to maximise instrumental accuracy. In principal, any use of direct observation should begin by validating the measuring instruments. Observers must be trained, and the researcher must verify, at different moments, if each one is precise and stable, and if they all furnish essentially the same recordings of the same phenomena. This can be done by filming one or more scenes and making each observer observe the scene, independently and repeatedly.

Instrumental reliability may be estimated in a number of ways. First of all, parametric models of variance analysis, and of multiple regression may be applied to determine the degree of error introduced by a single observer at different stages of observation, or during different repetitions of the same observations, or to estimate the degree of error introduced by two independent observers (Cohen, 1968; Cronbach et al., 1972; Kerlinger and Pedhazur, 1973). Most of the time, observational techniques do not produce measures which can be placed in a scale of intervals. Only multiple regression analysis may be applied to this type of measure, by using dummy variables. The use of other techniques may also be indicated, namely, logistic regression (Cox, 1970) and analysis of multidimensional tables of contingencies (Bishop et al., 1975; Colgan and Smith, 1978; Cox, 1970). Some of these techniques, however, postulate that each element of the recording being analysed, or each category of elements, is independent of the others. Units sampled by direct observation, are, more often than not, interrelated. In fact, one of the reasons motivating the use of observation is precisely the desire to establish the relations between various types of behaviour. We must nevertheless assure ourselves, before applying a given technique, that the basic postulates governing its use are respected.

Once he or she verified the basic validity of his, her measuring instruments, a researcher most often no longer has either the financial means or the time to repeat his, her study of instrumental accuracy. He or she usually contents himself, herself with a regular measurement of intra- and inter-observer agreement, compared with those determined during the initial study. This indicates if each observer works as a stable instrument for the entire duration of the research project. Still, despite a current belief, the measure of inter-observer agreement is not a measure of instrumental accuracy (Herbert and Attridge, 1975). It does not indicate to which point the observers are accurate, unless their recordings are compared to a pre-established standard, nor does it provide a clue as to the degree of stability present in the recording of observations, unless it is taken at many different moments over the duration of the research project. It is perfectly possible to obtain a very high degree of agreement between two observers, despite a very low degree of reliability when precision and instrumental stability are considered. All observers may, effectively, apply the same incorrect interpretations of the same definitions, may make the same mistaken observations, may simultaneously be subject to the same degree of betterment or worsening of the exactitude of their measurements.

Some of the ten or so indexes which serve to verify the stability and agreement between observers are presented in Table 8. Hollenbeck (1978) has done critical analysis of these indexes. The choice of index depends upon the scale of measurement used. The percentage of inter-judge agreement, based on the relations between the number of agreements and the total number of agreements and disagreements between two observers is strongly advised against, as it does not take the number of cases attributable to chance into account. However, it is very much still in use. On the other hand, two indexes seem to be particularly well-recommended, namely, Kappa, for nominal data, and Robinson's A (1957), for data on a scale of interval. 

The Kappa statistic, developed by Cohen (1960), advantageously replaces the percentage of agreement so frequently used. Kappa makes corrections for agreements attributable to chance: Kappa = (Po - Pc)/(1 - Pc), Po being the proportion of observed agreements and Pc that of agreements due to chance. An example will illustrate the application of Kappa. Table 9 presents a matrix simulating the agreements between two observers who viewed, independently, a video tape showing agonistic exchanges between two fish. This manner of arranging agreements in a table is especially practical during observer training. It permits rapid detection of the sources of disagreement and confusion over various units. Agreements for the six types of behaviour are placed diagonally; the disagreements, on either side, are easily located. In Table 9 therefore, there seems to be perfect agreement concerning the behaviour Flight; on the other hand, there is partial disagreement for the units Attack and Bite and more disagreement over the units Lateral display and Tailbeat. First of all, the proportions relative to the lines and columns are calculated, and their cross products added (Pc). The cross-product of the two proportions determines the associated probability of agreements due to chance, for the given behavioural units. In relation to the diagonal, the proportion of observed agreements (Po) is calculated. The complete formula is, finally, easily applied. 

Table 9. Matrix of agreements between two fictious observers looking at the same events. For instance, both observers agreed on 7 occasions, both stating that an attack had just occurred; they disagreed on 5 other occasions, observer 1 stating that an Attack had occurred while observer 2 reported the occurrence of a Bite instead. 

Other properties of Kappa may be of interest. This statistic is accompanied by a sampling theory (Fleiss et al., 1969); its distribution is known and it is relatively easy to interpret. This index was modified by Fleiss (1971) to adapt to simultaneous agreements between more than two observers. It is also possible to weigh this index in function of the consequences of certain disagreements (Cohen, 1972). 

As for Robinson's A, it applies to rating scales. For example, two judges may rate, on a scale of one to five, the intensity of an identical response or property in different subjects. While the correlation between the two series of ratings is based on the existence of a linear relation, the agreement associated with A necessitates a perfect match of the two sets of ratings.

 

APPENDIX

 

Definition of agonistic and epigamic behaviour units in the Green Swordtail fish (Xiphophorus helleri, Pisces, Poeciliidae)

agonistic: related to the establishment and maintenance of dominant-subordinate relationships.

epigamic: related to fecondation of internal eggs in Xiphophorus (which is an ovo-viviparous species)

GENERAL

Approach: slow movement of a fish toward his or her conspecific. It is initiated from a distance exceeding approximately twice the total length of the initiator.

Flutter: the fish is oriented and placed perpendicularly against one of the sides of the aquarium and swims laterally or up and down as if trying to get out of the aquarium. 

AGONISTIC METACATEGORY

Menacing supra-category:

Lateral-display: consists of the spreading of all fins, with the exception of the ventral fins which are kept close to the body. The body presents, most often, a typical sigmoid posture (S-drohen). The orientation of the individual is usually perpendicular to and in front of the conspecific, or parallel to it, when both fish are mutually displaying.

Tail-beat: the initiator adopts lateral-display posture, curves his body sharply, and rhythmically beats with his tail at a frequency of 2-10 units per minute. Tail-beats are administered in the direction of the opponent, as if water currents were aimed at his head, flank or tail.

Offensive supra-category:

Attack: sudden acceleration of an individual toward his conspecific, the initial distance being at least equal to the length of the initiator.

This unit may be followed by a bite.

Butt: butting harshly with the tip of the jaws on the front or on the side of the head of a rival.

Bite: grasping, with the mouth, such anatomical parts of the other individual as the ventral or pectoral fins, gonopodium or even the mouth of the opponent (mouth-fight).

Defensive supra-category:

Flight: is associated to an approach or attack or bite or lateral-display or tail-beat unit emitted by one individual. The threatened fish rapidly avoids the charging opponent or escapes his presence by a sudden acceleration separating both fish. This fleeing is sometimes accompanied by a submissive posture.

Submissive posture: lowering of the dorsal and caudal fins and spreading of the ventral ones. It is sometimes accompanied by tilting on the side. The defensive unit is an interactive behavioural pattern since it entails the concomitant initiation of an aggressive act by the opponent.

To hide: To hide behind the siphon or filter.

Piquet: holding a vertical tail-down posture while keeping immobile, usually in one corner of the aquarium or behind the siphon.

Interactionnal agonistic units:

Chase: the initiation of either one attack, bite, nip, lateral-display, tail-beat or approach by a given individual, and its temporal association with a defensive behaviour, or flight by the opponent. 

Mutual offense: simultaneous or quasi-simultaneous attacking or biting. This includes circling and mouth-fight.

Mutual menace: simultaneous or quasi-simultaneous initiation of lateral-spread display or tailbeat.

EPIGAMIC METACATEGORY

Sexual following and surveillance: The male approaches the female from behind, hovers, and/or follows her as she usually responds by accelerating or by moving a short distance forward and then stopping.

Sexual butting: The male (more rarely the female) butts gently with the tip of his jaws (nibbles) either in the vent region, usually behind or between the pelvic fins, or on the front or the side of the head.

Sexual waltz: Epigamic display in the male. It consists of the male positioning himself in front of the female, as if presenting a barrier to further forward or sideward movement of the female, and/or then "waltzing" up, down, sideways, forwards or inversely so as to maintain his station, transverse to the female's head (blocking).

It may facultatively back up rapidly, sideways to her, bringing his body to lie side by side with that of the female (gliding). These movements are so rapid as to be difficult to follow with the eye. The presence of one of these elements (blocking and/or gliding) is noted as "waltz". 

Copulation attempt: Usually following an epigamic display, the male swings his gonopodium forward and to the side nearest the female and then rolls his body away from the female at an angle of about 20 degrees. From this position, he makes forceful forward thrusts, aiming at the vent of the female. 

Copulation: The male and female fish hover motionless with the gonopodium in the genital aperture for at least one second. 

Dotting: The female responds to approach by the male by moving a short distance forward, not more than twice her own total length, and then stops.

 

REFERENCES

ADAMS, R.M., & MARKLEY, R.P. (1978). Assessment of the accuracy of point and one-zero sampling techniques by computer simulation. Report presented at the Animal Behavior Society Annual Meeting. Seattle, Washington.

ALTMANN, J. (1974). Observational study of behavior: sampling method. Behaviour, 49, 227-265.

BAERENDS, G.P., DRENT, R.H., GLASS, P. & GROENEWOLD, H. (1970). An ethological analysis of incubation behaviour in the herring gull. Behaviour, suppl., 17, 135-235.

BEAUGRAND, J.P. (1984). Observation directe du comportement, in: Robert, M. (Ed.), Fondements et étapes de la recherche scientifique en psychologie (2e édition), (pp. 167-218). Edisem. 

BEAUGRAND, J.P., CARON, J., & COMEAU, L. (1984). Social organization of small heterosexual groups of green swordtails (Xipho^horus helleri, Pisces, Poeciliidae) under conditions of captivity. Behaviour, 91, 24-60.

BEAUGRAND, J.P. & ZAYAN, R. (1985). An experimental model of aggressive dominance inXiphophorus helleri (Pisces, Pœciliidæ). Behavioural Processes, 10, 1-52. 

BICKMAN, L. La récolte des données I: Les méthodes d'observation

BISHOP, Y.M.M., FIENBERG, S.E. & HOLLAND, P.W. (1975). Discrete multivariate analysis: theory and practice, M.I.T. Press, Cambridge, Massachusetts.

BLURTON JONES, N.G. (1972). Ethological studies of child's behaviour, Cambridge University Press.

BOICE, R.: Observational skills. Psychological Bulletin, 93, 3-29, 1983.

BUNGE, M.(1980). The mind-body problem: a psychobiological approach. Pergamon.

CAIRNS, R.B. (1979). The analysis of social interactions: methods, issues, and illustrations. Wiley.

CHRISTENSEN, L.A. (1977). The negative subjects: myth, reality or a prior experimental experience effect, Journal of personality and social psychology, 35, 392-400.

CHRISTENSEN, L.A. (1980). Experimental methodology (2e ed.). Allyn & Bacon.

CHRONBACH, L.J., GLESER, G.C., NANDA, H., & RAJARATNAM, N. (1972). The dependability of behavioral measurements: theory of generalizability for scores and profiles.Wiley. 

COHEN, J. (1960). A coefficient of agreement for nominal scales, Educational and psychological measurement, 20, 37-46.

COHEN, J. (1968). Multiple regression as a general data analytic system. Psychological Bulletin, 70, 426-443.

COHEN, J. (1972). Weighted chi square: an extension of the kappa method, Educational and psychological measurement, 32, 61-74.

COLGAN, P.W., SMITH, J.T. (1978). Multidimensional contingency table analysis, in Colgan, P.W. (Ed.), Quantitative Ethology, pp. 145-174. Wiley.

COX, D.R. (1970). The analysis of binary data. Methuen.

COX, D.R. & LEWIS, P.A.W. (1966). The statistical analysis of series of events, Methuen.

DANE, B. & VAN DER KLOOT, W.G. (1964). An analysis of the display of the goldeneye duck (Bucephala clangula L.), Behaviour, 22, 282-328.

DANTZER, R. & MORMEDE, P. (1979). Le stress en élevage intensif. Masson..

DERMAN, C., L.J. GLESER, & I. OLKIN: A guide to probability theory and application. Holt, Rinehart & Winston, New York, 1973.

DORÉ, F., LEFEBVRE, L., CAVANAGH, P. & JOLY, R. (1977). Decision-making in observational studies: the Erlangian distribution. 

Report presented at the American Psychological Association, San Francisco.

DRUMMOND, H. (1981). The nature and description of behavior patterns, in: Bateson, P.P.G. & P.H. Klopfer (Eds), Perspectives in ethology, Vol.4, 1-33, Plenum.

DUNBAR, R.I.M. (1976). Some aspects of research design and their implications in the observational study of behaviour, Behaviour, 58, 78-89.

FAGEN, R.M. & YOUNG, D.Y. (1978). Temporal patterns of behaviors: durations, intervals, latencies, and sequences, in: Colgan, P.W. (Ed.), Quantitative Ethology: 79-114, Wiley.

FARINA, A.J. & WHEATON, G.R. (1973). Development of a taxonomy of human performance: The task characteristics approach to performance prediction. JSAS Catalog of selected Documents in Psychology, 3, 26-27 (Ms No. 323).

FISKE, D.W. (1980).Unwarranted optimism. American psychologist, 35, 935-936.

FLEISHMAN, E.A. & QUAINTANCE, M.K. (1984). Taxonomies of Human Performance: the description of human tasks, Academic Press.

FLEISS, J.I. (1971). Measuring nominal scale agreement among many raters. Psychological bulletin, 76, 378-382, 1971.

FLEISS, J.I., J. COHEN, & B.S. EVERETT: Large sample standards errors of kappa and weighted kappa. Psychological bulletin, 72, 323-327, 1969.

GELLERT, E.: Systematic observation: A method in child study, Harvard educational review, 25, 179-195, 1955.

GOLANI, I.: Homeostatic motor processes in mammalian interactions: a choreography of display, in: Bateson, P.P.G. & P.H. Klopfer (Eds.), Perspectives in ethology, Vol. 2: 69-134, Plenum, New York, 1976.

GUILFORD, J.P.: The nature of human intelligence. McGraw Hill, New York, 1967.

HAZLETT, B.A., & W.H. BOSSERT: A statistical analysis of the aggressive communications systems of some hermit crabs. Animal behaviour, 13: 357-373, 1965.

HERBERT, J., & C. ATTRIDGE: A guide for developpers and users of observational systems and manuals. American educational research journal, 12: 1-20, 1975.

HINDE, R.A.: Animal behaviour: a synthesis of ethology and comparative psychology. McGraw Hill, New York, 1970.

HOLLENBECK, A.R.: Problems of reliability in observational research, in: Sackett, G.P. (Ed.), Observing behavior, Vol. 2: Data collection and analysis methods: 79-98, University Park Press, Baltimore, 1978.

HUTT, S.J., & C. HUTT: Direct observation and measurement of behavior, Charles C. Thomas, Springfield, Illinois, 1970.

HYLAND, M.: Introduction to theoretical psychology, MacMillan Press, London, 1981.

KERLINGER, F.N.: Foundations of behavioral research (2e éd.), Holt, Rinehart &Winston, New York, 1973.

KERLINGER, F.N., & E.J. PEDHAZUR: Multiple regression in behavioral research, Holt, Rinehart & Winston, New York, 1973.

KILEY-WORTHINGTON, M.: Behavioral problems of farm animals. Oriel Press, Londres, 1977.

KIMBALL, J.P. : The formal theory of grammar. Prentice-Hall, Englewood Cliffs, New Jersey, 1973.

KIRK, R.E.: Experimental design: procedures for the behavioral sciences. Brooks/Cole, Belmont, Californie, 1968.

KRUSKAL, J.B.: Multidimensional scaling and other methods for discovering structure, in: EINSLEIN, K., A. RALSTON & N.S. WILF (Eds.). Statistical methods for digital computers, Vol. 3: 296-339, Wiley, New York, 1977.

LAW, A.M., & W.D. KELTON: Simulation, modeling and analysis, McGraw-Hill, New York, 1982.

LEFEBVRE, L.: L'organisation temporelle du toilettage chez la perruche ondulée Melopsittacus undulatus, Thèse de doctorat inédite, Université de Montréal, 1978.

LEHNER, P.N.: Handbook of ethological methods, Garland Press, New York, 1979.

LEMON, R.E., & C. CHATFIELD: Organisation of song in cardinals. Animal behaviour, 19: 1-17, 1971.

LEVINE, J.M., & W.H. TEICHNER: Development of a taxonomy of human performance: an information-theoretical approach. JSAS Catalog of Selected Documents in Psychology, 3, 28 (Ms No 325), 1973.

LORENZ, K.Z.: The foundations of ethology, Springer Verlag, New York, 1981.

MACCORQUODALE, K., & P.E. MEEHL: On a distinction between hypothetical constructs and intervening variables, Psychological review, 55, 95-107, 1948.

MACHLIS, L.: An analysis of the temporal patterning of pecking in chicks, Behaviour, 63: 1-70, 1977.

McGREW, W.C.: An ethological study of children's behavior. Academic Press, New York, 1972.

NAUR, P.: Report on the algorithmic language ALGOL 60. Journal of the association for computing machinery, 6: 1-17, 1963.

MILLER, R.B.: Task taxonomy: Science or technology ? in W.T. Singleton, R.S. Easterly, & D.C. Whitefields (Eds), The human operator in complex systems, Taylor & Francis, London, 1967.

OLLASON, J.C., & P.J.B. SLATER: Changes in the behaviours of the male zebra finch during a twelve hour day, Animal behavior, 21: 191-196, 1973.

PRUSCHA, H., & H. MAURUS: Classifying agonistic behavior patterns according to their function in the communication process. Contemporary primatology, 1975: 245-253, 1975.

ROBINSON, W.S.: The statistical measurement of agreement. American sociological review, 22: 17-25, 1957.

ROSENTHAL, R.: How often are our numbers wrong? American psychologist, 33: 1005-1007, 1978.

SACKETT, G.P.: Observing behavior, Vol. 2: Data collection and analysis methods, University Park Press, Baltimore, 1978.

SCHNEIRLA, T.C.: The relation between observation and experimentation in the field study of behavior, Annals of the New York Academy of Sciences, 51: 1022-1044, 1950.

SLATER, P.J.B. (1974). Bouts and gaps in the behaviour of zebra finches, with special reference to preening, Revue du comportement animal, 8: 47-61, 1974.

SLATER, P.J.B. & LESTER, N.P. (1982). Minimising errors in splitting behaviour into bouts. Behaviour, 79, 153-161.

STRAYER, F., & STRAYER, J. (1976). An ethological analysis of social agonism and dominance relations among preschool children. Child development, 47, 980-989, 1976.

THINES, G., & A. LEMPEREUR: Dictionnaire général des sciences humaines, Louvain-la-Neuve (Belgique), CIACO, 1984.

TINBERGEN, N.: Social behaviour in animals. Wiley, New York, 1953.

TUOMELA, R.: Theoretical concepts in neobehavioristic theories, in: M. Bunge (Ed.), The methodological unity of Science, 123-152, Dordrecht-Holland, D. Reidel, 1973.

WATSON, J.B. (1908). The behavior of noody and sooty terns. Carnegie Institution of Washington, 2 (publication no 103), 187-255.

ZAYAN, R.C. (1981). Problèmes et recherches de l'éthologie appliquée. Réflexions sur l'autonomie et l'interdisciplinarité d'une nouvelle science. Unpublished Report, Centre Albert Michotte de Biologie du Comportement, Université de Louvain (Belgium).

© Jacques P. BEAUGRAND 

maitto:beaugrand.jacques@uqam.ca