Data collection methods

As illustrated by Figure 3.1 (which illustrates the overall mixed-method methodology – in the previous section 3.1), this section presents the methods used for collecting the data for each research question.

3.2.1 Literature review

Data from which to derive the elements of the WCF were text excerpts, passages and quotations which were selected because they contained information about impacts, guidelines and recommendations, as well as animal’s needs, characteristics, activities, and environments as discussed by welfarists and biotelemetrists (reported in section 2.2 and 2.3). We acquired the relevant material by reviewing publications in the literature.

3.2.2 Collaborative requirements workshops

The WCF was proposed as a tool for facilitating a requirements analysis for wearability. In order to do so, the framework was delivered to teams of designer contributors who participated in three collaborative requirements workshops for establishing wearability requirements for a GPS tag consistent with cat’s characteristics and needs (details are in section 6.1.3).

Growingly popular among interaction designers (Preece et al. 2015, p. 363), collaborative requirements workshops are variants of the focus group technique and are employed as an ad-hoc method for requirements elicitation. As the name ‘requirements workshops’ indicates, they are specifically tailored to carry out an early requirements activity (Gottesdiener 2003). While focus groups are typically employed to evaluate products that are already on the market, whereby participants are usually asked to express how they use an own device or give their opinions about a product of interest (Cooper et al. 2014, p. 53; Preece et al. 2015, p. 237), requirements workshops are specifically employed to “quickly elicit, prioritise, and agree on” a common set of requirements based on which to start a design process (Gottesdiener 2003).

According to Gottesdiener (2003), in order to have a successful requirements activity during a workshop, the following is key:

1) The involvement of all representative stakeholders related to the project (i.e. project sponsors, product champions, direct and indirect users, advisors, and suppliers), since they may have different outlooks and needs that may translate into contrasting requirements.

2) A careful schedule for the team activities and a structure set up on deliverables that helps the participants to focus on the task.

3) A collaborative environment which is conducive to addressing the concerns of all participants.

4) The presence of a facilitator who follows the schedule and leads the teamwork. This is necessary both to explain the deliverables and team activities included in the plan and to supervise on the collaborative aspects of the workshop, checking that the participants agree on the requirements, and ensuring that the shiest ideas are given voice.

However, we contravened the first guideline by grouping representatives of the same stakeholder category together, since the objective of our workshops was to evaluate whether the WCF is a useful tool for eliciting wearability rather than carry out a product development. For this, it was more useful to have teams whose members had relatively homogeneous backgrounds, which would allow us to better identify any background- dependent differences in the use and usefulness of the framework, and whether teams having different backgrounds and interests were reaching similar outcomes. Hence, each team was composed by specialists of similar background, interest in technology, and knowledge about the cat species, while across teams there were diversity of such aspects. In respect to the second, third, and fourth guideline, we adhered more literally to Gottesdiener’s model. Thus, in our workshops, we arranged and timed the workshop activities, and administered the WCF as the central deliverable (second guideline); we facilitated a collaborative environment by keeping the number of contributors for each workshop low (four, to be precise) and gathering together people acquainted each other (these were deemed as fair measures to have an equal exchange of opinion among participants and minimise the main drawback of the technique, that is having the loudest participants impose their opinions) (third guideline); we kept the function of the facilitator as proposed by Gottesdiener (fourth guideline) while adding an additional function: the facilitator was also a cat expert who could be consulted whenever the participants had doubts about the biology of the species. Due to this double function, the facilitator trained herself to keep the two roles clear and separate (Preece et al. 2015, p. 255).

The workshops produced a data set in the form of transcripts of dialogues among designers. We firstly read them to derive wearability requirements heuristically (section 6.1.4) that were used to design a prototype during a prototyping stage of the research (section 7.1.2); secondly, we conducted a thematic analysis of the data (described in section 8.2) to evaluate whether the WCF was a useful tool for designers to account for the characteristics and needs of animal stakeholders.

3.2.3 Ethological observations in wearability studies

In order to evaluate the WCF usefulness, the requirements established heuristically through the WCF needed to be validated and improvements of the wearability of the prototype (derived from the use of the WCF) needed to be assessed. This was done by involving animals to establish requirements empirically and to gauge, through their reactions, their experience with devices. Two wearability tests, one with two off-the-shelf GPS tags (study 1) and the other with the prototype derived from the workshop activities (study 2), were

carried out in order to gain insights into the wearability of the different devices from an animal’s perspective (details are in sections 5.4 and 5.5 for study 1, and in section 7.2 for study 2).

To investigate the cat participants’ WX with the devices we chose to apply an ethological approach. Ethology is a science dedicated to the understanding of animals’ behaviour which makes use of repeated field observation to characterise and quantify specific behaviours (Tinbergen 1963).

Field observation is a common method for collecting behavioural data both in Human- Computer Interaction (HCI) and Ethology. However, physical, sensory and cognitive interspecies differences between human investigator and animal studied (resulting in differences of perception and communication barriers) underpin key differences in how observation is approached in the two fields. In HCI, direct observation in the field is usually employed as a descriptive qualitative method to understand the details of what users do in naturalistic settings (Preece et al. 2015, pp. 252-254); and usually observational field studies significantly differ from quantitative observation methods applied in laboratory settings, where participants operate under controlled conditions. Either way, a key aspect is that in HCI, observation methods can be complemented by self-reporting methods to somewhat validate observational data. This is of course not possible when working with non-human participants.

In ethology, observation is approached by choosing and defining behavioural categories to observe and quantitatively measure behavioural parameters (e.g. frequency, duration, latency, intensity) in naturalistic settings (Martin and Bateson 1993, pp. 62-66). Observational data are then usually (but not necessarily) treated statistically to verify hypotheses on the meaning of observed behaviours, which thus emerge from quantitative data (e.g. measuring the roaring rate in red deer stags and correlate it with fight occurrences to test whether the roar is an indicator of fitness (Clutton-Brock et al. 1979)). This enables ethologists to interpret the meaning of animal behaviour in a relatively objective way, thus reducing the risk of anthropomorphic interpretations (Martin and Bateson 1993, p. 18; Tinbergen 1963). However, when a single episode of a salient behaviour shows a clear effect, ethologists describe it in a qualitative way. Hence, the observational method adopted by ethologists involves qualitative and quantitative observations in non-manipulative experiments in natural and non-controlled settings (i.e. in the field), in which behaviours of interest are described and measured to test hypotheses on why the described behaviours might occur (i.e. exploring their possible meaning). In this regard, salient or unexpected behaviours are annotated and described qualitatively as a basis for prospective measurements, though quantification is performed to achieve as objective as possible an interpretation of animals’ behaviour (Dawkins 2007).

A key challenge of interpreting animals’ behaviours in the field is that researchers cannot control the environment and actions of the studied individuals as they could in a laboratory setting. Indeed, in the wild, individuals may disappear from the sight of the observer thus interrupting the recording, or they may respond to the observer’s presence in a way that invalidates the data (e.g. if they express curiosity, escape or otherwise alter their behaviour); external stimuli such as temperature, light, or interactions with conspecifics are non- controllable variables that may confound the observation of a possible cause-effect relation, and animals are observed as they carry on with their daily routines. These issues are addressed through an observational technique that focuses on a) controlling the observer instead of controlling the animals and their environments; and b) applying an experimental methodology to naturalistic observations (Grafen and Hails 2002; Dawkins 2007).

The method of ethological observation is grounded in a framework of four choices that the observer has to make, while observing three principles of experimental research. When planning an observation, the observer needs to choose (Dawkins 2007, pp. 73-88):

1- The ‘level of observation’ – whether to observe individual animals, or groups, or body parts. For example, when testing whether herds of sheep protect their off-springs by keeping them surrounded by the adults during transhumance, the observation level is the group, such as familial units in the herd; while when studying if off-spring always move when their parents move, the observation level is the individual young);

2- The ‘unit of behaviour’ – namely, the exact behavioural pattern to be observed (e.g. the approach of young towards their parents);

3- The sampling technique – which depends on what is available to observe, and what question the observer is trying to answer (e.g. when studying a rare behaviour, this should be recorded every time it is observed; when studying recurring behaviours, it might be more convenient to establish fixed periods of time and record them only when they are happening within those periods);

4- The type of record - whether the chosen behaviour has to be registered continuously (exact start and end each time it is performed; e.g. how long it takes to the young to get close to their parents) or as a point in time (e.g. just the occurrence).

The three principles of experimental research (Grafen and Hails 2002, pp. 76-85) require the researcher:

a) to perform independent replications in order to treat data statistically and therefore detect whether events occurred by chance;

b) to randomise the independent variable in order not to confound the many dependent variables that naturalistic environments present; and

c) to remove variation by comparing any manipulated conditions with either a control condition or a control group.

Ethological observation applied to this research

Given its quantitative and qualitative characteristics, ethological observation is potentially an effective tool for identifying meaningful reactions that might be caused by the presence of a biotelemetry device on the animal’s body. In particular, the controlled measuring of selected behaviours over time is especially suitable when it is uncertain which environmental variables might cause the animal’s reactions (and thus when the observed behaviours might be easily misinterpreted if observed solely qualitatively). At the same time, the method also allows the observations of salient self-explanatory incidents to be analysed qualitatively. Since we wished to understand the WX of cats while they were wearing tracking devices, behaviours hypothesised as indicating discomfort were counted (see section 5.2, pp. 78-80), while overt behaviours toward the tag clearly showing discomfort were qualitatively described. This exploratory-and-experimental combination was particularly important for this research since it allowed us to interpret the cats’ behaviours both in a generalised and individual fashion, and thus overcome the issues of misleading correlation (Dawkins 2007, p. 8) while appropriately accounting for individual differences (Feaver et al. 1986). Specifically, when the ethological protocol relies on statistical analysis for the interpretation, behaviour tends to be considered meaningful if its occurrence is statistically significant. In this respect, animals that show reactions different from the unit of behaviours chosen would be considered outliers and their behaviour dismissed as non-significant (Feaver et al. 1986). When evaluating the responses of individual animals to technological interventions, statistical analysis provides for reliability and validity of the behavioural interpretation (e.g. that a behaviour indicates discomfort). However, some individual behaviours that fall outside the hypothesised reactions might be especially meaningful, particularly if they are clearly directed at the device. For example, if an animal was to chew off components of the device but was the only individual of the sample population to do so and only did it once, this important behaviour might be omitted from the overall analysis or treated as an anomaly. However, for the purposes of designing animal-centred technology, that anomaly may indicate a very noteworthy design flaw. Hence, integrating the quantitative observational protocol with descriptive observations strengthens the methodology, especially when this is applied to the design context.

Although ethological observations are conventionally employed to study animals in the wild without intervening in their daily life, in order to measure the effects of tracking devices on animal wearers, we needed to include an environmental manipulation in the experimental design: fitting animal participants with tracking devices or prototypes. Once fitted with the devices, animals were observed conducting their habitual activities in their habitual environment, according to the conventional observation protocol. This approach allowed us

to measure and interpret the cats’ behaviour in a relatively objective and reliable way, while minimally disrupting the participants’ daily life. This was particularly important on both scientific and welfare grounds, since cats are commonly known to be prone to stress if their habits and environment get changed (Rochlitz 2005).