Domain identification.

Domain identification involves clearly specifying the purpose, boundaries, and a priori dimensions of the construct of interest [7]. This is especially important for a novel, primarily theoretical concept like duty of care. As such, we conducted a narrative literature review to build on the preliminary conceptual descriptions of duty of care provided in our previous paper [4].

During this literature review, we found that our original conceptual descriptions of duty of care resonated strongly with Schwartz’s Norm Activation Model (NAM) [8]. The NAM is a psychological model of prosocial behaviour which centres around feelings of moral obligation as the primary motivator. As such, we have adopted this model as a framework for examining the potential dimensions and mechanisms of duty of care and how they may work together to motivate behaviour.

While originally developed for explaining altruistic behaviours, the Norm Activation Model has since been applied to other prosocial behaviours and areas such as pro-environmental behaviours [9], transport choices [10], energy saving [11], recycling [12], and occupational safety [13]. It has often been used in conjunction with the Theory of Planned Behaviour [6], to increase its predictive ability when studying behaviour with a prosocial element e.g., [11, 14–16]. Indeed, Schwartz argued that the model was not exclusive to altruistic behaviour and should apply in any circumstance where behaviour is motivated by moral obligations [17]. In the current context, dog owner care and management behaviour can be considered prosocial in that it is concerned with the welfare of another. However, that is not to say that all owner behaviours are motivated by prosocial intentions. Indeed, the same behaviour may arise from different motivations in different people, or even a combination of motivations within the same person. For example, dog walking may be motivated by prosocial intentions of providing for the welfare of the dog, but it may also be motivated by self-serving motivations for exercise, or pragmatic motivations that if the dog isn’t taken for a walk it will become destructive. As such, by examining duty of care in the framework of the NAM, we do not suggest that all dog management behaviour is solely motivated by duty of care. Rather, we aim to understand the role it may have in different behaviours relative to other potential motivators. While obligations or duties towards dog walking have been investigated in two studies [18, 19], to our knowledge the full NAM has yet to be applied to the care and management of companion animals.

Schwartz broadly outlines the NAM in terms of three basic propositions:

1. Obligations: prosocial behaviour is influenced by the strength of an individual’s feelings of moral/personal obligation to perform that behaviour
2. Activation: situational factors ‘activate’ an individual’s ‘personal norms’ in a specific circumstance, generating these feelings of moral obligation
3. Defence: action stemming from feelings of moral obligations may be neutralised by defences regarding the relevance or appropriateness of the obligation.

Beyond these central propositions, the interpretation and practical application of the NAM has been rather inconsistent. There are 5 main inconsistencies we have seen (and which have been identified to differing degrees by others previously [e.g., 20, 21]. In the interest of brevity, we will provide a single example of each:

1. inconsistency in the definitions of key variables (i.e., variables are called the same thing but defined differently), e.g., de Groot and Steg [22] define “awareness of consequences” as “whether someone is aware of the negative consequences for others or for other things one values when not acting prosocially.” (pg 427) while Harland et al. [20] define it as “a person’s receptivity to situational cues of need” (pg 324).
2. inconsistency in the naming of key variables (i.e., variable is defined in the same way but called something different), e.g., what de Groot and Steg [22] defined as “awareness of consequences”, is called “problem awareness” in Steg and de Groot [21].
3. varied use of behaviour-specific versus more general level beliefs, e.g., Onwezen et al. [23] used general beliefs related to “protecting the environment” while Bamberg et al. [10] used more specific beliefs around car use and public transport.
4. different number of variables are used with some key omissions, e.g., Harland et al. [20] use seven of the NAM variables originally outlined by Schwartz, while most studies only use three or four [e.g., 24].
5. the modelled relationship between variables differs, i.e., some use a mediation model [22], some use a moderation model [25], and others use a combination of moderation and mediation [23].

It is beyond the scope of this paper to present a detailed analysis of these inconsistencies. As such, in our interpretation of the NAM we have drawn primarily on the original descriptions in Schwartz [8] and Schwartz and Howard [26]. Fig 3 provides a summary of the 4 key stages in the NAM [21] and the associated variables within these steps. Here we have purposely not included arrows to denote the structure of the model as this is an ongoing debate and an empirical question to be addressed later.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. The Norm Activation Model.

Key stages and associated situation-specific state variables derived from Schwartz [8]. Trans-situational traits coloured yellow.

https://doi.org/10.1371/journal.pone.0285278.g003

In short, the process begins with the activation stage which moves through several steps including the initial awareness or recognition of a problem or party in need, followed by the identification of actions that could relieve that need (efficacy), belief in one’s ability to perform those actions, and finally the sense of responsibility to be involved with the needy party. These steps lead to the activation of personal norms (defined later) and anticipated emotions which generate feelings of moral obligation to perform the behaviour. Once the feelings of moral obligation have been generated, the associated action and its outcomes are assessed in terms of costs and benefits to the actor. Where costs outweigh the benefits, a range of defence mechanisms can be employed to neutralise the feelings of moral obligation.

In the original NAM, the primary outcome variable is the behaviour of interest. However, where we have placed duty of care in the hypothesised PCC model would suggest that the outcome from the NAM variables would be the behaviour-specific attitudes of the TPB. This may be an oversimplification and requires empirical testing. It may well have a unique influence on behaviour directly, bypassing the TPB variables. However, the place of duty of care within the PCC model should not affect the present goal of developing a scale to measure it.

We will now briefly define the key NAM variables as these will constitute the hypothesised a priori dimensions of the duty of care scale [7]. While activation is the first step in the model sequentially, we will begin with Personal Norms as they are the cognitive foundation of feelings of moral obligations and thus, central to the model.

Obligation stage. Personal norms are an individual’s beliefs or self-expectations about how they ought to behave in a particular situation. Differing from the more commonly used ‘social norms’ which are the expectations or standards of others for how we should behave, personal norms are our own personal expectations for behaviour. Of course, these self-expectations originate in social learning, but over time they are internalised and modified in each individual [8]. Personal norms also stem from values, but differ from them in that they are situation/behaviour-specific while values are trans-situational and more enduring (for more detail on how personal norms differ to and complement values, see section 3.1 of Schwartz and Howard [26, pg 238]).

Where social norms motivate behaviour through social sanctions, personal norms motivate behaviour through sanctions related to the self-concept and associated self-evaluative emotions [8, 10]. That is, a person may be motivated to behave in line with their personal norms to avoid negative self-evaluation and feelings of guilt, failure, or loss of self-esteem, or to gain or maintain feelings associated with positive self-evaluation like pride, security, or enhanced self-esteem. Consequently, these internalised norms are self-reinforcing, personal standards for behaviour [8] and as such, are particularly promising with regards to behaviour change and the sustainability of that change. Additionally, this feature of personal norms whereby actions are evaluated in terms of the self-concept and the moral implications for one’s sense of self is what differentiates them from other attitudinal constructs (such as those included in the Theory of Planned Behaviour), which make evaluations based on material, social, and psychological payoffs [26]. With regards to the basic propositions of the NAM, personal norms are the cognitive foundation for feelings of moral obligation (the primary motivator in the NAM). That is, feelings of moral obligation are the product of activated personal norms combined with anticipated self-evaluations/emotions [8]. Furthermore, from an operational perspective, it is difficult to measure the feeling of moral obligation that motivates behaviour in specific situations; it is more feasible to measure the salient beliefs these feelings derive from (i.e., personal norms) which are potentially more stable [8]. This is a similar approach to that commonly employed when operationalising the Theory of Planned Behaviour, measuring salient beliefs as opposed to attitudes directly.

Activation stage. Awareness of need and problem awareness are concerned with the activation of personal norms and have been used somewhat interchangeably in NAM studies. They refer to the extent to which a person is aware of the need of the focal entity, whether that be another person, an animal, the environment etc. While conceptually similar, in practice, awareness of need is more often used with regards to more specific needs (e.g., reducing car use to reduce climate change), while problem awareness has been used more generally (e.g., climate change as a major issue of concern). Additionally, it is worth noting that several studies, particularly within the environmental domain, have used the term “awareness of consequences” to describe awareness of need or problem awareness. However, in Schwartz’s original presentation of these concepts, awareness of consequences is described as a trait-like variable (described later), not a situational activator [20].

Schwartz [8] highlights that the relationship between awareness of need, personal norms, and behaviour can be influenced by several factors. Initially, to reach the threshold of basic awareness, the prominence and clarity of the need are important, as well as the individual’s personal receptivity to the need cues. In the case of dog care, receptivity to need cues would also include understanding/knowledge of what a need cue is e.g., understanding dog behaviour and body language that is indicative of a problem for the dog. Indeed, anecdotally, many dog owners are receptive to some need cues like problem behaviours (e.g., they notice that their dog is destroying furniture), but aren’t aware that they are a sign of compromised welfare or may misconstrue them (e.g., they consider the dog is being spiteful). Once the threshold of basic awareness of need is reached, the intensity or seriousness of the need then also has an impact on behaviour. That is, if we consider the behaviour of taking the dog to the vet, action is more likely to occur in response to more intense need (e.g., dog with broken leg) than less intense need (e.g., dog with skin condition). Stimulus overload is also likely to influence awareness of need by reducing attentiveness to need cues. That is, owners that have “a lot going on”, perhaps with work or with other family commitments, are likely to not be as attentive to the needs of their dog and miss some of these need cues. In these situations, the time available to interact with the dog and therefore time available to notice need cues would also be impacted. These examples highlight some of the wide range of potential factors that can influence an owner’s awareness of need or problem awareness.

Efficacy and outcome efficacy are also terms used interchangeably in the literature. Once a need has been recognised efficacy refers to the perception/recognition/identification of potential actions that could relieve the need.

Following from efficacy, ability is the extent to which the person believes they can perform the identified actions to relieve the need. In this way it is similar to perceived behavioural control from the Theory of Planned Behaviour. This is a key area for neutralising obligations later in the process through denial of ability.

Ascription of responsibility/situational responsibility is often described as the acceptance of responsibility for the consequences of performing or not performing the identified behaviour/s. However, Schwartz originally described it as “a sense of connection or relatedness with the person in need” and “the sense of some responsibility to become involved with the needy” [8, pg 246]. He also differentiated responsibility from obligation in that: “Feelings of obligation are directed toward the performance of specific acts; their strength is a function both of the connection with the needy (responsibility) and of the implications of the act for self-evaluation.” Existing role relationships (i.e., relationships with reciprocal role expectations of each party) automatically implicate a sense of responsibility and this is likely the case for many dog owners. However, there are several other sources of responsibility, the most relevant for dog ownership being the concept of dependency. As such, the greater the owner’s perception of the dog’s dependency on them, presumably the greater their ascribed responsibility. On the contrary, if an owner perceives their dog and its welfare to be relatively independent of the owner, their acceptance of responsibility is likely to be lower and perhaps even facilitate a denial of responsibility.

Awareness of consequences has been perhaps the most commonly confused concept from the NAM, yet is also one of the most commonly used. In the original descriptions, awareness of consequences is described as a general predisposition or trait-like “tendency to become aware of the consequences of one’s behaviour for others” [8, pg 229] or the tendency to “perceive situations in terms of the consequences their own behaviour has for others” [8, pg 225]. It is trans-situational and as such, it facilitates, but is not the same as the situational awareness of need described above. It was originally measured using a general scale, not specific to the behaviour of interest [8, 20].

Defence stage. Where the costs of action outweigh the benefits, defence mechanisms may be employed to neutralise feelings of moral obligation. Each of the original situational activators are potential targets for defence against action. That is, one may deny the reality or severity of the need or problem, deny their ability to perform the actions required, or deny their responsibility in the particular situation.

In addition to the situational defences described above, responsibility denial has been identified as a trans-situational trait likely to influence this process. Like awareness of consequences, it is a general predisposition to deny responsibility for the consequences of one’s actions [8]. There is a 28 item set that is designed to measure this across settings [8].

Care beliefs and standard of care. While the NAM provides a good framework for the moral obligation component of duty of care, there is one other element to be considered: the level of care that is implied by the obligation. That is, different owners may have similar feelings of moral obligation to provide care for their dog, but the actual standard of care they perceive to be appropriate may differ. These differences could arise from differences in value-based judgments and differences in knowledge about the needs or capacities of dogs. For example, one owner may consider that an appropriate level of care for their dog is for it to sleep inside on a bed, while another may consider it more appropriate for dogs to sleep outside in a more “natural” state.

Model structure. As mentioned above, different studies have used different statistical models of the NAM, typically either a moderation or a mediation model. A review by de Groot and Steg [22] of the use of the NAM in the context of environmental work, argued that the mediator model is more appropriate. However, this review only modelled three NAM variables (personal norms, awareness of consequences [awareness of need/problem awareness], and ascription of responsibility). The mediator model suggests that a person must be aware of the problem or need before feeling responsible for it [23]. Additionally, they tested a range of models to determine the appropriate place of emotions, finding that they best served as mediators between personal norms and intentions, but with personal norms also having a direct effect on intentions as well. Few studies have used efficacy, ability, and defence in their NAM models, so it is unknown how they should interact with the other elements. However, we consider it likely that they work in similar ways to emotions and as such, potentially have a mediating effect within the activation stage and the norm-obligation-action transition. Fig 4 provides a graphical depiction of the hypothesised model structure.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Hypothesised model of NAM elements for use in the duty of care scale.

Relationships supported by previous research depicted by solid arrows. Postulated relationships in dashed lines.

https://doi.org/10.1371/journal.pone.0285278.g004

Evaluation of existing instruments: Is a new scale needed?. There are many instruments currently available for assessing various characteristics of human-animal interactions. Indeed, Wilson and Netting [27] identified 140 different tools that had been developed up until 2010. Since then, many more have been added to this list. However, there is presently no standardised, psychometrically validated scale to evaluate the various concepts related to duty of care as described here. Other studies have typically only used a small number of items (one-three) to measure obligation [18, 19]. However, given the proposed multidimensional nature of duty of care we consider that a more extensive scale is required.

Deductive item generation: Literature review and existing scales.

Existing scales regarding the Human-Animal Relationship were examined for items relevant to the dimensions outlined in the domain identification. Additionally, studies that utilised the NAM in other contexts were drawn on to structure other items not previously used in the human-animal literature.

One important consideration was the level of behavioural specificity at which the items would be focussed. It is well recognised that the more specific the items, the better they are at predicting individual behaviours (i.e., single act criteria) [28]. However, when examining a general concept such as appropriate dog care, there are a multitude of individual behaviours that constitute this. One potential solution is to utilise the concept of a multiple act criterion [29]. That is, measuring general attitudes towards dog care is more likely to be predictive of a set of related behaviours in a single behavioural domain with the same target (e.g., the dog) [8]. By examining a range of specific actions, a class of behaviour is made which is more likely to be predicted by a general attitudinal antecedent. As such, in initial item generation, we included both items regarding general dog care and welfare as well as specific management behaviours to cover both single and multiple act criteria. The specific management behaviours used were those that were considered more discretionary (e.g., spending time, exercise, preventative health care) as they would likely have greater variability in related attitudes than foundational care behaviours (e.g., feeding, providing water).

Inductive item generation: Interviews.

To elicit the “real-world” views of dog owners, semi-structured qualitative interviews were conducted with adult (18yr+) dog owners. Self-selection bias is a major challenge when studying pet owner attitudes as individuals who are more engaged and interested in dogs are more likely to volunteer for dog-related research. As such, we used several measures to reduce this bias and to capture a broader range of views. Recruitment was conducted in two ways. Firstly, posts were made in Facebook community groups, but these were limited to non-dog related groups e.g., “Buy, Swap, Sell” or Community Noticeboard pages. Secondly, the Royal Society for the Prevention of Cruelty to Animals (RSPCA) Victoria (Australia) assisted by emailing the study advertisement to people who had engaged with their “Latrobe Loves Pets” community outreach program. The Latrobe Loves Pets program provided support (e.g., assistance with veterinary bills, free pet food or supplies) to individuals who were facing challenges providing care for their animals. In doing so, we hoped to capture parts of the community that may face barriers to providing optimal pet care. An incentive of a $50 visa gift card was provided for each participant to both encourage participation by people who are not dog-enthusiasts and to compensate participants for their time. Additionally, we used a brief screening questionnaire in the Expression of Interest (EOI) form which focussed on beliefs about dogs and key management practices (see S1 File). This enabled us to select participants with varied views from a wider pool of EOIs. A total of 363 EOIs were received.

Thirteen (n = 13) interviews were conducted via video call (Zoom Video Communications Inc, San Jose, CA, USA) in September 2020, January 2021, and February 2021. Participants included 4 males and 9 females ranging in age from 25–54, living in various areas of Victoria, Australia including inner city Melbourne, suburban Melbourne, and regional/rural Victoria. Participants also varied in their household characteristics including families with young children, families with adult children (left home), couples, singles, and singles living in share housing. The dog-owning experience of participants also varied; some had extensive experience caring for dogs while others were first-time dog owners. Each interview lasted for approximately 1 hr and all interviews were conducted by the first author (CRG). To begin each session, a quick-response sentence completion exercise was conducted to elicit “top of mind” responses to key concepts and to facilitate further conversation. The remainder of the interview was semi-structured, addressing themes of general beliefs about animals and their own pets, general motivation for management activities, responsibility, perceived control/barriers, emotions, and finally, duty of care (full interview protocol available in S2 File). Duty of care was deliberately left until the end of the interview to see if it would be mentioned by participants unprompted earlier in the interview.

The Zoom session for each interview was recorded and fully transcribed for thematic analysis. Transcripts were manually coded using NVivo Version 12 (QSR International Pty Ltd. 2018) using a combination of inductive and deductive coding. While some themes/nodes derived from the domain identification and background theory, others were developed in response to the data. All nodes were reviewed, consolidated, and amended in an iterative process. Questionnaire items were then developed to reflect these key themes (e.g., I brought my dog into my household so I have a moral obligation to care for them). Additionally, some items were derived directly from statements made by interviewees (e.g., I owe my dog fun times).

Initial item pool.

A full list of the generated items can be found in the (S3 File) with their relevant source. The majority of items were developed in a Likert style using a 7-point response scale from ‘strongly disagree’ to ‘strongly agree’.

For two of the dimensions (ascription of responsibility and problem awareness/awareness of need), two different approaches to their measurement were developed for testing and comparison. For ascription of responsibility, the first set of items were more general “agree-disagree” items (e.g., I feel completely responsible for my dog’s safety). The second set focussed on more specific states of the dog (i.e., the consequences of care and management) and asked participants “To what extent do you feel responsible for the following:”, then used piped text to insert the dog’s name into the individual items e.g., “[dog’s name] being happy”. The response scale was a 5-point scale from “not at all responsible” to “completely responsible”. For problem awareness/awareness of need, both approaches used the same response scale (agree-disagree) but the first set was more general in nature (e.g., My dog could be happier) while the second set pertained to the specific management or interactive behaviours (e.g., If my dog does not get enough exercise their welfare will suffer).

The three items developed for efficacy could also be interpreted as ability (or perceived behavioural control). That is, in saying “There are things I could do to improve my dog’s life” it includes both the identification of behaviours required to improve the dog’s life and a judgement of whether the person could do them. As such, we combined these elements of the NAM for the present scale. Importantly, when used in the context of the PCC model, additional items would be included to assess perceived behavioural control under the Theory of Planned Behaviour section, and general knowledge, which would also add to this. Hence the few items (three) used here are justified.

Online survey.

This stage of scale development involves delivering the full set of generated items (S3 File) to a large sample of the target audience to facilitate factor analysis and item reduction. To do so, an online survey was constructed using Qualtrics software, Version April 2021 (Qualtrics, Provo, Utah, USA). The survey comprised 3 main sections and 8 subsections based on the dimensions identified in domain identification.

1. Demographics (to ensure adequate representation of different demographics)
2. Duty of Care scale items (see S3 File)
   1. Personal norms (duty beliefs)
   2. Ascription of responsibility
   3. Problem awareness and awareness of need
   4. Efficacy/Ability
   5. Emotions
   6. Standard of care
3. Existing scales (for testing construct validity)
   1. Lexington Attachment to Pets Scale
   2. Monash Dog Owner Relationship Scale

Demographic questions included gender, age, locality (urban, suburban, regional), number of dogs in the household, dog’s name, source of dog acquisition, and age at acquisition. The dog’s name was recorded as this was used in later questions to elicit a more contextualised response than simply referring to “your dog” (e.g., “to what extent do you feel responsible for [Dog’s name]’s mental state”).

At the end of each page (subsection) there was an optional text box for participants to provide feedback on the items (e.g., comprehension and clarity) or to indicate if there was some aspect of that concept that had not been covered by the items.

The Lexington Attachment to Pets Scale (LAPS) [30] and Monash Dog Owner Relationship Scale (MDORS) [31] are both existing, psychometrically validated scales and were included in the survey to investigate construct validity (see Stage 3: scale evaluation).

The survey was disseminated through social media, predominantly Facebook and Twitter. As with the interview recruitment, posts were made to primarily non-dog related Facebook community groups e.g., ‘Buy, Swap, Sell’ or ‘Community Noticeboard’ pages. Data collection occurred from June 8, 2021 to July 20, 2021.

Exploratory factor analysis.

Each subsection of the duty of care scale was analysed separately. This began with examining descriptive statistics for each item, inter-item correlations, and the qualitative feedback on the items (provided in the text boxes at the end of each section) to determine which items should be included in the subsequent exploratory factor analysis (EFA). Items with low inter-item correlations (<|0.3|) [7] and substantial negative qualitative feedback were excluded from factor analysis. Where possible, only items that did not demonstrate significant skew (skewness <|1|) were included [32]. However, for some subscales (e.g., duty beliefs), this was not possible as most items in the subscale demonstrated skew. As such, the principal axis factoring (PAF) extraction method was used for the EFA as it is the most robust method for skewed variables [33]. Factors were extracted based on both eigenvalues and the scree plot. Small coefficients (absolute value less than 0.33) were suppressed [34, p. 649] and variables that did not load were removed before reanalysing the smaller set of items. Rotation was not required as all but one analysis yielded a single factor solution.

After initial EFA solutions were found, the internal consistency of each factor was tested using Cronbach’s alpha (α). If any items could be removed to improve internal consistency they would have been, however this was not the case for any of the scales. Factor scores were saved using the regression method and descriptive statistics were conducted to examine their distribution. These analyses were performed with IBM SPSS Statistics for Windows, version 27 (IBM Corp., Armonk, NY, USA).

Structural equation modelling.

Following the EFA, zero-order correlations were conducted on the factor scores to provide a preliminary understanding of the relationships between the different factors. Subsequently, structural equation modelling (SEM), combining both confirmatory factor analysis (CFA) and path analysis, was conducted to examine the structure and fit of the full model. SEM was performed using IBM SPSS Amos for Windows, version 27.0.0 (Amos Development Corp., Wexford, PA, USA). While the PAF extraction technique was used for the original EFA as it is most appropriate for skewed variables, maximum likelihood (ML) is the only method available in Amos for SEM. As such, to ensure this was appropriate for the SEM we reanalysed the original data with the maximum likelihood method which produced very similar results to the PAF method. The hypothesised model (Fig 4) was used as a basis for the SEM with modification indices inspected to determine if improvements in model fit could be achieved.

Scale scores.

Once the structure and items were confirmed, simple scale scores were calculated for each subscale by summing the values for each relevant item. These were then compared to the factor scores using correlations. This is an important step to simplify future use of the scale, assessing the validity of using simple scale scores as opposed to factor scores.

Construct validity.

The relationships between the duty of care constructs and the LAPS and MDORS scales were assessed for construct validity. The concepts of attachment to pets (LAPS) and the dog-owner relationship (MDORS) were expected to correlate with the duty of care constructs, yet remain distinct. In this way, these scales were used to assess both convergent and discriminant validity.

The MDORS and LAPS each have three validated sub-scales. Scores for each subscale were calculated by summing their relevant items. Additionally, an overall score was calculated for each scale by summing all items. Zero-order correlations were conducted between the MDORS and LAPS scores and the duty of care simple scale scores.

Personal norms (duty beliefs).

All items were significantly skewed with the exception of “I feel obligated to provide my dog with as close to a ‘natural life’ as possible”. However, there were several comments in the feedback about this item being confusing. Therefore, it is likely that the greater spread for this item is more reflective of variation in interpretation of the question as opposed to actual differences in underlying beliefs. As such, this item was removed prior to factor analysis. Additionally, there were many comments objecting to the use of the phrasing to “owe” dogs, so all items using this term were removed. A single factor solution comprising 7 items was found, accounting for 35.8% of the total variance and with an acceptable level of internal consistency (α = .78) (Table 2). Factor scores remained significantly skewed (skew = -1.82, SE = 0.11).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Item loadings for exploratory factor analysis of personal norms (duty beliefs) dimension of the duty of care subscale.

https://doi.org/10.1371/journal.pone.0285278.t002

Given that the resulting factor scores remained skewed, we considered rewording the items to use more moderate language, making the items less likely to elicit “strongly agree” responses. Reworded items were trialled on a new group of participants as part of a separate study (n = 88). All reworded items were skewed except two items (all reworded items available in S4 File). The least skewed item was “My duty to care for my dog is. . .” with a 7-point response scale from “the least important thing in my life” to "the most important thing in my life” (skew = -.58, SE = .26). The second asked "How personally obligated do you feel to minimise your dog’s negative experiences in life?” with a 7-point response scale from “Not obligated at all” to “Completely obligated” (skew = -.88, SE = .26). Factor analysis and SEM could not be performed on this sample as there was not a sufficient sample size.

Ascription of responsibility.

The first approach to measuring ascription of responsibility (general; agree-disagree) demonstrated less variability, lower inter-item correlations, and more skew than the second approach (specific states; “How responsible would you feel…”). Many of the individual items used for the second approach were normally distributed. Highly skewed items were removed before all remaining items from both sets were pooled together for the EFA. A 10-item, single factor solution was found accounting for 47.44% of the variance with a good level of internal consistency (α = .87) (Table 3). Despite the low loading of item AR1_14 (-.332), we considered it important to retain this due to its unique semantic content, reflecting an element of denial of responsibility. Interestingly, the most skewed items (that were removed prior to factor analysis) often related to positive states (e.g., level of responsibility felt for dog being happy, healthy, or safe).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Item loadings for exploratory factor analysis of “ascription of responsibility” dimension of the duty of care subscale.

https://doi.org/10.1371/journal.pone.0285278.t003

Problem awareness and awareness of need.

In contrast to the ascription of responsibility items, the results of the general set were less skewed than those of the specific set. After removing items based on skew and inter-item correlations, both sets were pooled together for EFA. After an initial two factor solution was found, only the top six items loading on the second factor were retained to ensure a similar number of items for this section as the previous. This final 10-item, two factor solution accounted for 43.58% of the variance (Table 4). Factor 1 reflects the concept of “problem awareness” (PA) and was named as such. Factor 2 reflects an awareness of the consequences or impact of management behaviours on the dog’s behaviour and welfare. However, to avoid confusion with the broader trait-like variable “awareness of consequences” (see domain identification section), we will call this “awareness of impact” (AoI). Both factors exhibit acceptable-good internal reliability (α_PA = .84, α_AoI = .73).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Item loadings for exploratory factor analysis of “awareness of need” dimension of the duty of care scale.

https://doi.org/10.1371/journal.pone.0285278.t004

Efficacy/Ability.

These items were not significantly skewed, displayed a good spread of responses, and had good inter-item correlations. A single factor solution accounting for 59.96% of the variance was found with an internal consistency of α = .80 (Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Item loadings for exploratory factor analysis of “efficacy/ability” dimension of the duty of care scale.

https://doi.org/10.1371/journal.pone.0285278.t005

Emotions.

All items demonstrated good inter-item correlations and there was no negative qualitative feedback about any of the items. However, all items except two were highly skewed with little variation. Initially EFA was trialled with all items. A 3-factor solution was found to be the best fit for these data. However, when these factors were incorporated into the SEM there were issues with overly large regression weights (β>1). Consequently, we attempted to re-do the EFA excluding the most skewed items (skew>2). A reasonable two factor solution was found, however the distribution of the resultant factor scores for factor 1 were still heavily skewed (skew = -2.28) and the reliability was unacceptable (α = .66). Factor 2, despite consisting of only two items, demonstrated better reliability (α = .76) and was not skewed (skew = -.16). However, when trialled in the SEM, we found that it had no relationship with the duty beliefs (β = 0.01). As such, it was concluded that the items developed for assessing the emotional component of the NAM were inadequate and excluded from the model.

While the qualitative feedback did not highlight any issues with the items themselves, several participants felt the need to explain that they are more motivated by positive emotions (e.g., happy, rewarding) than by negative emotions (e.g., guilt, shame).

Care beliefs/standard of care.

Similarly to the emotions section, all items except two were highly skewed with little variation. A similar process to that described for emotions was conducted with similar results. The initial factor solution caused problems in the SEM and the revised EFA (excluding the most heavily skewed items) failed to produce acceptable factors. As such, this subsection was excluded from further analyses and SEM.