3 How To Build Trust

Within the literature, several models have been developed that seek to provide a systematic account of how trust is built, maintained and, in some cases, broken. Including these models here is helpful to situate the recommendations, as it highlights how trust dynamics function and reveals what features “go into” building trust.

Before responding to the question of how trust is built, it is important to be mindful that trust is not a static process, and as implied in stage models of trust, it changes over time (Rousseau et al., 1998; Korsgaard, 2018). Setting these stages at the forefront displays synergism with the recommendation to apply differentiated types of trust, and tailor bespoke recommendations. This recommendation of bespoke trust building efforts is important as, just like how not all types of trust are the same, how trust is built can vary depending on the different stages of trust. Recognising this helps situate the discussion around the relevant stage, type and model of trust, and ensures that a focused overview of the dynamics involved can be provided.

An overview of three stages is provided by Korsgaard (2018):

“The first, deterrence-based (Shapiro et al., 1992) or calculus-based (Lewicki & Bunker, 1996), refers to early-stage relationships that are relatively transactional. Trust is conditional on the potential costs and benefits to each party of trusting and being trustworthy.

In the second stage, knowledge-based trust is based on the accumulated knowledge of the partner’s trustworthiness over repeated interactions.

The third and most robust stage of trust is identification-based trust wherein the actor’s values and interests are aligned with the partner. As the basis of trust shifts, the paths of influence between cooperation and trust change over time.”

As the above synopsis makes clear, different stages align more closely with different definitions of trust. Namely, the first stage displays clear synergy with trust as a rational choice; the second emphasises reciprocity, learning and the importance of past experiences; and finally, stage three refers to what Faulkner (2018) may term affective trust, which is more intimate and values-based than the predictive alternative.

This recommendation to focus on the earlier stages of trust building does not mean that expressing shared values should be avoided, only that one should not be reliant on the processes outlined the final stage of trust building: firstly, because the type of trust that official statistics require does not necessitate that level of depth or intimacy; and secondly, because if statistical producers wait until stage three to engage in trust building behaviours, there is a very real risk that audiences will have already become distrustful. This is not unduly pessimistic. It is based on the ‘established wisdom that trust is built slowly and lost quickly’, alongside a concern that if opportunities to build trust are neglected, distrust may fill the void (Korsgaard, 2018, p.20).

With this conclusion in mind – that efforts to build trust in official statistics should be predominately concentrated in the first two stages of trust – this review provides a brief synopsis of four distinct routes to trust building. The four routes are based on models of trust which have been thematically grouped. The resulting techniques and mechanisms involved in building trust are: 1) learning from past experiences; 2) evaluating qualities and characteristics; 3) gaining knowledge and developing understanding; and finally, 4) relying on the context and borrowing trust from others. It is beyond the scope of this review to provide a detailed analysis of these four strategies. For this, readers can make use of the links and references provided to investigate the models further.

For the purpose of this review, this section provides a short, focused synopsis. Throughout the review, links which signpost to the relevant model have been included. This reverse signposting centralises the topic focus (i.e., trust in actor/object) and centres recommendations with a topical priority.

3.1 Learning from Past Experiences

This section includes trust models which emphasise repeated learning, experiences and evaluations of past performance as the mechanisms through which trust is built. These models are particularly useful in the earlier stages of trust building.

3.1.1 Experience and Familiarity (Luhmann, 1979)

This model of trust relies on the notion that trust is the product of repeated exposure to positive experiences. To quote Luhmann (1979, p.19–20), ‘familiarity is a precondition for trust and distrust.’ This model states that the outcome of trust (or in the case of negative experiences, distrust) will be decided primarily on the ‘assumption that the familiar will remain.’

To summarise the model: familiarity creates expectations of future outcomes. These expectations are based on past experiences. Consequently, if the experience is positive, trust will be increased; inversely, if the experience is negative, trust will be destroyed.

3.1.2 Performance–trust hypothesis (Yang & Holzer, 2006)

This model proposes that performance underpins trust. It suggests that observations of positive performance increase trust in the actor, whereas failing to meet the performance criteria can have a negative impact on trust.

To provide an example: ‘individuals are more likely to express trusting attitudes if they also assess government performance in a positive light’. This clearly involves a subjective assessment of what constitutes a ‘positive’ performance. This demonstrates that although appraisals of trust may differ from one person to another, even after witnessing the same performance, this does not invalidate the model, because the subjective assessment of performance may still determine trust on a personal level.

3.1.3 Universal Sequence for Trust (Dietz, 2011) and Distrust (Six & Latusek, 2023)

This model functions as a feedback loop whereby certain inputs (i.e., disposition, character and institutional context) inform beliefs, decisions and actions. The decision of whether to respond proactively (trust) or protectively (distrust) is based on feedback from actions which then inform the input.

As Six and Latusek (2023) explain, though this model is universal, it is not rigid. To elaborate: the model is flexible in the sense that it acknowledges that different stages of the process (input, beliefs, decision, action) may demand more weight depending on the given situation or context but that the process – that is, the sequence through which the outcome of trust or distrust is reached – remains consistent.

3.2 Evaluating Qualities and Characteristics

Methods included within this section propose that trust is built on the basis of evaluative judgements based on personality, qualities and characteristics. These features may differ in accordance with different professional standards, and they are particularly suited to building trust in the second stage, as judgement and expectations are relied upon ‘in lieu of their immediate experience’ (Korsgaard, 2018, p.20).

3.2.1 Ability, Benevolence and Integrity (ABI) Model (Mayer et al., 1995)

This model highlights three dimensions of trust: ability (competence to perform the task); benevolence (cares about them); and integrity (adheres to similar set of values). It suggests that when evaluating the agent’s trustworthiness, each of these aspects is considered. These are subjective judgements.

To paraphrase Hamm, Smidt and Mayer (2019), the model itself displays a high degree of parsimony (ability to capture the concept in the fewest number of variables); has considerable empirical evidence to support the validity of the variables; and is purposefully designed to be consistently applicable across contexts – even if, as mentioned, the relative importance of the three aspects may vary.

In reviewing the existing literature on distrust, Six and Latusek (2023) signpost an alternative model – incompetence, malevolence and deceit (IMD) – which has been proposed for measuring distrust. The model uses the subconstruct’s incompetence (the actor’s lack of ability, knowledge or expertise to accomplish the assigned task); malevolence (harmful intentions); and deceit (perceptions of the other person as dishonest, treacherous or fraudulent in their character).

3.2.2 Trust-as-Evaluation Model (Hardin, 2002)

This model proposes that trust is a judgement of competence (skill) and integrity (commitment to do no harm): A trusts B (who demonstrates integrity) to do X (because they have the skill to do so). This model has inspired a breadth of theoretical literature and empirical studies.

To flesh the components of this model out with an example, one may be trusted to water plants but not to mind a child. Other examples could highlight a particular profession or skillset – one may trust a builder to construct a garden wall but would probably be more hesitant to trust them to perform surgery or land a plane.

3.2.3 Subjective–Rational Evaluation (van der Meer, 2010)

According to this account, trust is the product of a rational process of evaluation, whereby the potential trustor subjectively appraises the object of trust against a four-part typology. The assessment of whether the actor meets each criterion is based on prior experience and ‘perceptions’ of the actor’s ability to meet expectations (van der Meer, 2010, p.532). As such, it is both subjective and rational.

The four criteria shaping this evaluation are: competence (ability which may be general or domain-specific); intrinsic care (benignity, shared values and a commonality of interest); accountability (external systems which moderate negative behaviour and facilitate accountability for wrongdoing); and finally, reliability (predictability, keeping promises and behaving as expected).

This typology is overtly relational, bringing together ‘the characteristics of the citizens, and the political system, as well as their interplay’ [emphasis in original] (van der Meer, 2010, p.519). Although initially designed to evaluate political trust, this actor/system and intrinsic/external ‘interplay’ is also useful in cases of institutional (i.e., the UK Statistics Authority) and system-level trust (the statistical system).

3.3 Gaining Knowledge and Developing Understanding

The model included under this heading prescribes knowledge as the foundation of trust, or to be more specific, the absence of knowledge and understanding as liable for distrust.

3.3.1 Knowledge Deficit Model (Miller, 1983)

This model is based on the idea that “plugging the knowledgeable gap” will lead to more-positive attitudes (Irwin, 2014, as cited in Taylor et al., 2023). The model argues that a lack of understanding leads to scepticism – a feature of distrust – and to mitigate against this, further knowledge should be sought. This knowledge–attitude nexus was originally developed for science communication, and it continues to be endorsed, in spite of considerable evidence that the objective of increasing knowledge is unlikely to change behaviour, with ‘condescending claims of “public ignorance” too often serv[ing] to further alienate key audiences.’ (Nisbet & Scheufele, 2009).

OSR’s own work concludes that this approach is unlikely to effectively build trust in isolation. This is considered in more detail in OSR’s report on statistical literacy, which recommends that the onus is placed on the producer to better meet the needs of their audiences, rather than blaming audiences for gaps in their knowledge and viewing it ‘as a deficit that needs to be fixed’ (OSR, 2023).

3.4 Relying on the Context and Borrowing Trust from Others

The model outlined here explains how trust can be borrowed and shared between individuals and/or organisations within the same network. The conditions under which trust is shared are not universal. As such, establishing a better picture of the network dynamics relevant to official statistics may help tailor the recommendations more effectively.

3.4.1 Networks and Chains of Trust (Buskens, 2002)

This model proposes that one’s position in a network can influence levels of trust, with the agent’s sanction potential (determined via interconnectivity within the network); the importance, and duration, of future interactions (learning effects, including information of previous abuses of trust); and higher density and stronger ties helping support trust (Buskens, 2002).

The model also points to the importance of ascertaining whether trust is shared via a chain network or non-chained exchanges. As outlined by Cook and Santana (2020), in the former, trust can disintegrate if one actor in the chain behaves in an untrustworthy manner (the chain of trust collapses). In the latter, the network is vulnerable to free-riding, whereby one actor’s positive reputation is treated as a proxy for the trustworthiness of the others engaged within the network.

The second insight relates to one’s position within the network. Cook and Santana (2020) have suggested that members of a network can borrow status from those close to them in their networks. Thus, a high-status individual may be more trusted than those on the periphery. This is a result of their network position, and their connections, rather than any exemplification of trustworthy behaviours. Conversely, those in a peripheral position have fewer contacts to influence and may trust less-trustworthy members out of ‘desperation and dependence’ (Cook et al., 2006).