Rather than construct lists of many different welfare indicators and give each of them the same weight, I argue that the assessment of animal welfare should be directed at answering two key questions: I) Are the animals healthy? 2) Do they have what they want? Behaviour has a major role in answering both. Behaviour is currently used to help answer the first question through its use in the clinical and pre-clinical assessment of pain, injury and disease, and potentially could have an even greater role, particularly if used in conjunction with new technology. Behaviour is also of crucial importance in gauging what animals want, most obviously in the use of choice and preference tests, but also through other methods that are particularly suitable for on-farm welfare assessment. These include quantitative observations of the spatial distribution of animals and of behavioural ‘indicators’ of what animals want, such as vocalisations.

Standard cages for laboratory animals are often small, minimalist and barren. Such cages can compromise animal welfare, indicating that there are welfare-based reasons for improving their designs. However, a second issue, that is, whether animals from standard laboratory housing and husbandry conditions provide valid research data, also indicates that cage designs and husbandry methods need to be improved. This paper reviews various influences of standard laboratory cage design and husbandry. These include their effects on the repeatability of studies, models of neuro-degenerative disease, sensory development, physiology, and behaviour, the effects of standard social housing and standard handling, and the effects of maternal experience on the responses of offspring. These studies show that the development and responses of animals from standard laboratory housing and husbandry conditions are often unrepresentative and idiosyncratic, indicating that data are likely to have reduced external validity. An underlying question is whether animals from standard, barren laboratory cages are ‘abnormal’ and therefore might not provide valid baseline data. In terms of animal welfare, these studies indicate that standard laboratory housing may sometimes be associated both with reduced welfare and with reduced benefits gained from research. It is suggested that in a similar manner to the use of production measures when assessing cages for production animals, laboratory cages could be assessed in terms of their suitability to provide valid research data.

The assumption that animals are conscious and capable of experiencing negative sensations and emotions is at the core of most people's concerns about animal welfare. Investigation of this central assumption should be one goal of animal welfare science. We argue that theory and techniques from cognitive science offer promising ways forward. Evidence for the existence of conscious and non-conscious cognitive processing in humans has inspired scientists to search for comparable processes in animals. In studies of metacognition and blindsight, some species show behaviour that has functional parallels with human conscious cognitive processing. Although unable to definitively answer the question of whether the animals are conscious, these studies provide fresh insights, and some could be adapted for domestic animals. They mark a departure from the search for cognitive complexity as an indicator of consciousness, which is based on questionable assumptions linking the two. Accurate assessment of animal emotion is crucial in animal welfare research, and cognitive science offers novel approaches that address some limitations of current measures. Knowledge of the relationship between cognition and emotion in humans generates a priori frameworks for interpreting traditional physiological and behavioural indicators of animal emotion, and provides new measures (eg cognitive bias) that gauge positive as well as negative emotions. Conditioning paradigms can be used to enable animals to indicate their emotional state through operant responses. Although evidence for animal consciousness and emotion will necessarily be indirect, insights from cognitive science promise further advances in our understanding of this fundamentally important area in animal welfare science.

The use of demand functions to identify the behaviours most important to animals has been advocated widely. The principle is to place increasing cost on the opportunity to perform several behaviours and subsequently to rank these behaviours according to the change in their performance as a function of cost; this change is described by the elasticity of the demand function. However, the method has been criticised for placing the animal in too artificial a setting. Firstly, the animal works repeatedly for short periods of access to a resource, which may interrupt bouts of behaviour; secondly, animals are tested in isolation, which may affect their motivation to perform the target behaviour; and, finally, assumptions regarding the effect of prior deprivation and reward duration on elasticity of demand need to be tested. This criticism, however, is important only if these factors do affect the elasticity of the resulting demand function. This paper reviews experiments that have developed methods to assess the importance of various behaviours to farm animals and that have tested the effect of social context, length of deprivation of a resource and reward duration. It is concluded that the elasticity of demand function may be used to assess the relative importance of various behaviours, but that it is important to make sure that the experimental set-up yields valid estimates of the elasticity of demand.

Responses to potential threats to welfare vary greatly between species. Even closely related animals often differ in their fear of humans and/or novelty; their behavioural responses to pain; and when captive, their overall welfare and the form and frequency of their stereotypies. Such species differences stimulate hypotheses about I) the way that responses to challenge co-vary with other biological traits; 2) the adaptive value of particular responses; and 3) the factors predicting responses to evolutionarily new scenarios, such as captivity. We illustrate how these ideas can be statistically tested with multi-species comparisons, and show how techniques such as the Comparative Analysis of Independent Contrasts can be used to control for any non-independence of data points caused by species’ relatedness. For each of the three types of hypothesis, we then provide several welfare-relevant examples including one that has been fully tested (respectively, the relationships between sociality and anti-predator behaviour in antelopes; predation pressure, foraging niche and neophobia in parrots; and home range size and stereotypy in carnivores). Ultimate explanations such as these, based on species’ ecology and evolutionary history, have great explanatory appeal. Species comparisons can also have great practical value, allowing the test of hypotheses that would be almost impossible to investigate experimentally, and generating principles that allow predictions about the welfare of similar unstudied species. Multi-species data, for example from the many taxa held in zoos, thus hold enormous potential for increasing the fundamental understanding of animal welfare.

Studies of the brain inform us about the cognitive abilities of animals and hence affect the extent to which animals of that species are respected. However, they can also tell us how an individual is likely to be perceiving, attending to, evaluating, coping with, enjoying, or disturbed by its environment, and so can give direct information about welfare. In studies of welfare, we are especially interested in how an individual feels. Since this depends upon high-level brain processing, we have to investigate brain function. Brain correlates of preferred social, sexual and parental situations include elevated oxytocin in the para-ventricular nucleus of the hypothalamus. Abnormal behaviour may have brain correlates, for example, high frequencies of stereotypy are associated with down-regulated μ and κ receptors and dopamine depletion in the frontal cortex. Such results help in evaluating the effects of treatment on welfare. Some brain changes, such as increased glucocorticoid receptors in the frontal lobes or increased activity in the amygdala, may be a sensitive indicator of perceived emergency. Active immunological defences lead to cytokine production in the brain, vagal nerve activity and sickness effects. Some aspects of brain function can be temporarily suppressed, for example, by opioids when there is severe pain, or permanently impaired, for example, in severely impoverished environments or during depression. Coping attempts or environmental impact can lead to injury to the brain, damage to hippocampal neurons, remodelling of dendrites in the hippocampus, or to other brain disorganisation. Brain measures can explain the nature and magnitude of many effects on welfare.

In recent years, consumer demand studies have been used to objectively quantify the value that captive animals place on environmental resources. Considerable progress has been made in the development of effective methodologies to assess what resources are valued under controlled experimental conditions, but few of the findings from these studies have been implemented in commercial husbandry systems. A major obstacle has been the need to maintain internal validity in experimental studies at the possible expense of external validity. Experimental subjects may be poor representatives of commercially kept animals, especially where there is no replication of the animal's experience of commercial physical and social environments. This paper will discuss means of increasing the external validity of consumer demand studies. Firstly, experimental animals should be of representative genotypes, reared under commercial conditions and housed in test apparatus with free access to all of the resources that are normally available under commercial husbandry conditions. As well as providing a long-term closed economy, this allows manipulation of potentially important environmental factors such as prior experience. Secondly, naturalistic tasks should, where possible, be used in preference to abstract operant tasks since they appear to be easier to train animals to associate with rewarding resources, therefore many individuals can be investigated efficiently. Furthermore, naturalistic tasks appear to be less prone to operant bias than artificial tasks and can provide robust measures of value in terms of the maximum price paid for access to resources. Finally, for group-housed species, social costs such as enduring higher stocking densities and/or competition for resources may be an effective means of investigating cost-benefit trade-offs under commercial conditions. For example, if animals work to evenly distribute themselves over available space, this social spacing priority can be used to investigate whether they will endure higher stocking densities in order to exploit particular resources.

We estimate that stereotypies are currently displayed by over 85 million farm, laboratory and zoo animals worldwide. This paper investigates their reliability as welfare indicators, by surveying studies relating stereotypy to other welfare measures and by analysing the mechanisms underlying this behaviour. Where data exist, most (approximately 68%) situations that cause/increase stereotypies also decrease welfare. Stereotypy-eliciting situations are thus likely to be poor for welfare, although exceptions exist. Within such an environment, however, most (approximately 60%) accounts link individual stereotypy performance with improved welfare (cf approximately 20% linking it with reduced welfare). Thus, in a sub-optimal environment, non-stereotyping or low-stereotyping individuals could well have the poorest welfare, although again exceptions exist. Examining the mechanisms underlying stereotypy performance, we discuss four processes that could account for these complex links between stereotypy and welfare. Beneficial consequences from performing the specific source-behaviour of the stereotypy (‘do-it-yourself enrichment’), or arising from repetition per se (‘mantra effects’), may ameliorate welfare in poor environments. In addition, stereotypies that have become centrally controlled (habit-like), or that arise from autistic-like changes in the control of all behaviour (perseveration), are likely to be unreliable indicators of current state because they can be elicited by, or persist in, circumstances that improve welfare. To refine the role of stereotypy in welfare assessment, we suggest the collection of specific additional data to reveal when any of these four processes is acting. Until such research increases our understanding, stereotypies should always be taken seriously as a warning sign of potential suffering, but never used as the sole index of welfare; non-stereotyping or low-stereotyping individuals should not be overlooked or assumed to be faring well; simple measures of frequency should not be used to compare stereotypies that differ in age, form, or the biological or experiential characteristics of the performing animal; enrichments that do not immediately reduce stereotypies should not be assumed failures with respect to welfare; and finally, stereotypies should not be reduced by means other than tackling their underlying motivations.

Our ability to assess pain in animals in clinical situations is slowly developing, but remains very limited. In order to develop appropriate pain scoring schemes, numerous practical problems need to be overcome. In addition, we need to appraise realistically our current poor state of knowledge. Development of new scoring systems must be coupled with the increased education and training of those responsible for pain management, so that both the assessment and the alleviation of pain are steadily improved.

The distress experienced by animals during the induction of unconsciousness remains one of the most important and yet overlooked aspects of effective methods of anaesthesia and euthanasia. Here we show that considerable differences exist in the aversive responses elicited by 12 common methods of inhalational anaesthesia and euthanasia in laboratory rats and mice. Carbon dioxide, either alone or in combination with oxygen or argon, was found to be highly aversive to both species. The least aversive agents were halothane in rats and enflurane in mice. Exposing these animals to carbon dioxide in any form, either for anaesthesia or for euthanasia, is likely to cause considerable pain and distress and is therefore unacceptable when efficient and more humane alternatives are readily available.

Humans interact with fish in a number of ways and the question of whether fish have the capacity to perceive pain and to suffer has recently attracted considerable attention in both scientific and public fora. Only very recently have neuroanatomical studies revealed that teleost fish possess similar pain-processing receptors to higher vertebrates. Research has also shown that fish neurophysiology and behaviour are altered in response to noxious stimulation. In the light of this evidence, and in combination with work illustrating the cognitive capacities of fish, it seems appropriate to respond to a recently published critique (Rose 2002) in which it is argued that it is not possible for fish to experience fear or pain and that, therefore, they cannot suffer. Whilst we agree with the author that fish are unlikely to perceive pain in the same way that humans do, we believe that currently available evidence indicates that fish have the capacity for pain perception and suffering. As such, it would seem timely to reflect on the implications of fish pain and suffering, and to consider what steps can be taken to ensure the welfare of the fish that we exploit.

It is not enough to study animal welfare; our responsibility is to promote it. To this end, we need to step out of our laboratories and develop robust protocols for assessing welfare in groups of animals on farms or in the laboratory. While these protocols must incorporate principles derived from detailed scientific study they will, in practice, need to be based on clinical observations and records that an assessor can acquire at a single visit. Such assessments must also be comprehensive: approaches based only on behaviour, or motivational state, or physical appearance, or performance records, can never tell the full story. The ‘Five Freedoms’ provide a comprehensive template that incorporates the different elements that define welfare state. However they only describe welfare at the time of observation. Protocols, for example, for on-farm assessment need to include measurements and records that provide evidence of long-term consequences of the quality of husbandry practices (eg body condition, chronic injuries). This paper reviews approaches developed at Bristol for the on-farm evaluation of welfare in dairy cows and free-range hens. The criteria used to assess welfare from clinical evidence were developed through expert consultation using the ‘Delphi’ review process. Experts were also asked to make value judgements as to the seriousness of different elements of poor welfare by indicating the point at which intervention would be necessary to resolve specific problems. This study identified a serious limitation of Quality Assurance schemes that seek to encompass many different elements of welfare into a single index that ranks overall welfare as acceptable or unacceptable. Specific farms had specific welfare problems and these required specific solutions.

Animal welfare assessment commonly involves behavioural and physiological measurements. Physiological measures have become increasingly sophisticated over the years, while behavioural measurements, for example duration or frequency, have changed little. Although these measures can undoubtedly contribute to our assessment of an animal's welfare status, a more complex analysis of behavioural sequences could potentially reveal additional and valuable information. One emerging methodology that could provide such information is fractal analysis, which calculates measures of complexity in continuous time series. Its previous application in medical physiology suggested that it could reveal ‘hidden’ information in a dataset beyond that identified by traditional analyses. Consequently, we asked if fractal analysis of behaviour might be a useful non-invasive measure of acute and chronic stress in laying hens and in pigs. Herein, we outline our work and briefly review some previous applications of fractal analysis to animal behaviour patterns. We successfully measured novel aspects of complexity in the behavioural organisation of hens and pigs and found that these were stress-sensitive in some circumstances. Although data collection is time consuming, the benefit of fractal analysis is that it can be applied to simple behavioural transitions, thereby reducing subjective interpretation to a minimum. Collectively, the work to date suggests that fractal analysis — by providing a novel measure of behavioural organisation — could have a role in animal welfare assessment. As a method for extracting extra information from behavioural data, fractal analysis should be more widely examined in animal welfare science.

Cross-institutional studies that combine non-invasive physiological measures of stress responses and the assessment of individual differences in behaviour and temperament have great potential as tools for assessing the well-being of zoo animals and for identifying key environmental stimuli relevant to well-being. In addition, such studies allow comparison of animals under a wide variety of conditions and enable researchers to obtain sufficiently large samples sizes for statistical data analyses. Faecal corticoid measurements, a method recently developed to monitor adrenal activity in wildlife and domestic species, can be obtained non-invasively as part of the normal husbandry routine. While basic techniques still need improvement, and interpretation of the acquired measures can be challenging, assessment of faecal corticoid concentrations can provide a useful indicator of stress responses under a variety of captive conditions. Here we report on three studies that illustrate this approach and the results that can be obtained. An on-going study reveals significant differences in the pattern of variability of faecal corticoid concentrations between polar bears that are reported by keepers to perform stereotypic behaviour and those that do not. In another study, faecal corticoid measures indicated that stress responses to certain extraneous noises might interfere with the breeding of Hawaiian honeycreepers in captivity. In a study of clouded leopards, higher faecal corticoid concentrations were measured when cats were kept on public display or near potential predators compared to individuals maintained off exhibit or in the absence of visible predators. The findings of an on-going experimental study suggest a causal relationship between the provision of additional hiding spaces and a decline in faecal corticoid concentrations in clouded leopards.

Choice tests are commonly used to measure animals’ preferences, and the results of such tests are used to make recommendations regarding animal husbandry. An implicit assumption underlying the majority of choice tests is that the preferences obtained are independent of the set of options available in the test. This follows from two assumptions about the mechanisms of choice: first, that animals use absolute evaluation mechanisms to assign value to options, and second, that the probability of choosing an option is proportional to the ratio between the value of that option and the sum of the values of the other options available. However, if either of these assumptions is incorrect then preferences can differ depending on the composition of the choice set. In support of this concern, evidence from foraging animals shows that preferences can change when a third, less preferred option is added to a binary choice. These findings have implications for the design and interpretation of choice tests.

The provision of advice on animal welfare is an important part of the work of scientists in applied ethology, neurophysiology, veterinary epidemiology and other disciplines. Those who request guidance often expect advice that will help them to make progress in difficult discussions. Scientists want to live up to these expectations, but it is also important for them to clarify any scientific limitations. They are normally aware of limits to their advice, but these limits are sometimes not explicitly stated. Using the phrase broadly, we call this kind of limitation ‘scientific uncertainty’. We distinguish between the following four types of uncertainty: I) Ontological uncertainty, relating to the existence of animal feelings and other states relevant for animal welfare. 2) Conceptual uncertainty, stemming from the fact that some of the concepts used in animal welfare science are value-laden if used outside a narrow scientific context. 3) Lack of scientific evidence, stemming from a lack of scientific data on the problem in question. 4) Uncertainty about priorities, relating to the practical conclusions to be drawn in a situation with an open-ended set of ethical and other practical considerations. Scientific uncertainty is unavoidable. It is therefore essential, when giving scientific advice, to state the assumptions on which the advice is based. This makes scientific advice more objective, but also of more limited value to those who do not share the underlying assumptions.

Ethical principles guide decisions about what we consider to be right or wrong proposals and actions, so that when value judgements are made these engage us in ethics. Ethical thinking is clearly relevant to decisions about the way we treat animals, and such thinking has led to the commonly held view that animal use by people is acceptable provided that such use is humane. The ethical requirement that we treat animals humanely means that we must minimise the harms we do to all animals in our care or under our control. Accordingly, we must assess what harms are done to the animals, how bad each harm is in terms of its intensity and duration, what methods are available or can be developed to minimise each harm, and the relative effectiveness of those methods of harm minimisation. We must also seek to use the most humane methods currently available that can be practically applied. Ethically driven scientific evaluations of this sort have improved the humaneness of livestock slaughter and vertebrate pest control. For instance, detailed studies of different pre-slaughter stunning methods have validated their use to pre-empt the pain and distress that otherwise conscious animals would experience during the fatal neck cut and the short period of consciousness (sensibility) that follows it Likewise, reducing injuries caused by restraint traps has improved the humaneness of this vertebrate pest control method, and comparing the effects of different poisons has allowed the least humane ones to be identified. Difficult questions nevertheless remain at the problematic interface between quantitative scientific observations and their interpretation regarding the suffering that animals may experience, as well as questions about the relativities of different types of suffering.

In choosing among alternative wildlife management techniques, trade-offs between animal welfare and conservation, utility or economics are often apparent. This paper reviews the roles of science, scientists, regulators and educators in attempts to overcome this inter-dependence and to make simultaneous progress on all fronts. Illustrations are drawn in particular from trapping and pest population control. Against the real progress that can undoubtedly be made through scientific study, there is a regrettable lack of structure because of poor philosophical and logical coherence within the welfare movement.

Scientists recently attracted considerable public attention when they presented a featherless chicken tailored for production in hot climates. Although this chicken was actually the result of traditional breeding, it is an example of what might be achieved if targeted gene manipulation techniques become widely applied in agriculture. Through interfering directly with an animal's genome, scientists hope to be able to create animals with exactly the desired characteristics, such as lean meat or temperature tolerance. Industry and geneticists may be enthusiastic about the possibility of producing pork with polyunsaturated fatty acids or high-yielding dairy cows to be kept in tropical climates, but the European public often reacts with alarm at these prospects. A consistent pattern of the surveys conducted among members of the European public is that, of all of the potential biotechnology applications, those involving animals are the ones that people find the least acceptable. People fear a development of techniques that may get out of control, and they also have ethical concerns about humans’ right to ‘play God’ and about the welfare of the animals involved. All of these aspects seem to be relevant for an ethical discussion about animal biotechnology. Animal welfare scientists can play an important role by providing information for an animal welfare risk assessment at an early stage of research projects that involve the genetic modification of animals, and also by helping to develop guidelines for the housing and husbandry of animals with special needs. On the other hand, ethical problems remain that lie outside the area of science. In this paper we discuss the role of animal welfare science in aiding ethics decisions about animal biotechnology. We give a summary of the different ethical concerns expressed by ethicists and by the general public. Focusing on one of them, animal welfare, we give an introduction to the animal welfare implications of recent developments in reproductive and gene technologies. The importance of animal welfare aspects is discussed in relation to other ethical concerns about animal biotechnology.