Log in / Register
Home arrow Engineering arrow Measuring Electronic Word-of-Mouth Effectiveness: Developing and Applying the eWOM Trust Scale

Identification Stage

Specification of Content Domain, Construct Definition and Dimensionality

According to the majority of scholars, the first step in the scale development process is a sound theoretical definition of the focal concept, which should entail what is included in the domain of the construct and what is not (Netemeyer et al., 2003). Further, it shall be clarified to which entity the construct generally applies, the conceptual theme shall be outlined and a theoretical justification about the construct’s dimensionality should be provided (e.g., MacKenzie et al., 2011; Rossiter, 2002). While all these topics have been discussed in greater detail in Chapter 3, this sub-chapter is dedicated to an issue that may need further explanation. It’s the construct’s dimensionality or - more specifically - the relationship among the various first-order subdimensions (types of trust), as well as between the dimensions and the focal, second-order construct (eWOM trust).

In social sciences and specifically applied marketing research, constructs are regularly conceptualized and operationalized as multi-dimensional entities (Brewer, 2007; Lin et al., 2005). A construct is theorized to be multi-dimensional “when it consists of a number of interrelated attributes or dimensions and exists in multi-dimensional domains. In contrast to a set of interrelated uni-dimensional constructs, the dimensions of a multi-dimensional construct can be conceptualized under an overall abstraction, and it is theoretically meaningful and parsimonious to use this overall abstraction as a representation of the dimensions” (Law et al., 1998, p. 741).

Overview of the Research Process

Figure 4: Overview of the Research Process

It is also this thesis’ approach, as outlined earlier, to understand eWOM trust to be multidimensional, as a person’s generalized trust is likely to be manifested in multiple aspects. More specifically, eWOM trust is viewed as a second-order construct that is mirrored by three types of trust, namely (i) the consumer’s confident beliefs that the information conveyed in customer reviews is honest, useful, and benevolent (i.e., eWOM trusting beliefs); (ii) the consumer’s positive attitude towards eWOM in general (i.e., eWOM trusting attitude); and (iii) the consumer’s generalized willingness to rely on review information and recommendations for making purchase-related decisions. Further, the construct of eWOM trust is regarded as being - at the same time - multi-faceted, best described by a set of distinct aspects including cognitive, emotional and conative elements. This is also captured by the above mentioned sub-dimensions, where emotions, cognitions and conations trigger different types of trust. While each of these five facets can be understood as a separate sub-construct, at a more abstract level, they are all critical elements of a person’s generalized trust in OCR. This conceptualization takes on the tripartite conceptualization of a person’s mental orientation towards an object, advanced, for instance, by Bagozzi et al. (1979); but it at the same time parallels MacKenzie et al.’s (2011, p. 301) assertion of multi-dimensional concepts that “if the essential characteristics [i.e., dimensions] describe relatively unique aspects of the construct, eliminating any of [its dimensions] would restrict the conceptual domain of the construct, then the construct is multidimensional from a conceptual perspective”.

Given that eWOM trust is conceptualized to be a higher-order construct, the nature of the relationships within the construct has to be considered. Or, in other words, a critical issue when developing a new scale is to determine (i) the nature of the relationship between second-order latent construct and its sub-dimensions; as well as (ii) the relationship between the latent subconstructs and their manifest indicators (e.g., Diamantopoulos et al., 2008; Jarvis et al., 2003). Latent constructs can be defined as “phenomena of theoretical interest which cannot be directly observed and have to be assessed by manifest measures which are observable” (Diamantopoulos et al., 2008, p. 1204). On each abstraction level of the construct, two basic kinds of relationships are possible. Here, researchers typically make a distinction on the basis of whether the indicators on the item-level or the sub-dimensions on the construct-level are influenced by a latent first or respectively the second-order variable - or vice versa (Bollen, 2011). If the direction of the relationship is from the latent variable to its measures, these measures are regularly called reflective or effect indicators. In contrast, if the direction of the relationship is reversed - from the measures to the construct - the measures are commonly denoted as formative or causal indicators (Blalock, 1964; Bollen & Lennox, 1991).

The existence of this first kind of relationship (i.e., the reflective measurement model) has long been recognized in social sciences. Academic assumptions about this relationship are here based on classical test theory (Lord & Novick, 1968). Diamantopoulos and Winklhofer (2001, p. 269), for instance, state that “this perspective reflects the conventional wisdom on measurement [...], which is largely based on classical test theory and, in particular, the domainsampling model”. This theory asserts that variations in the indicators constitute effects (or manifestations) of variations in an underlying latent construct (Bollen & Lennox, 1991). Hence, reflective indicators depend on the latent construct or rather are caused by them (Diamantopoulos et al., 2008). An important characteristic of this kind of relationship is that changes in the latent variable lead to changes in all of its measures at the same time. Additionally, it corresponds to the general notion of the reflective measurement model that all measures have to be positively correlated (Bollen, 1984); therefore, the measures are more or less exchangeable.

The second kind of relationship (i.e., the formative measurement model) has only recently experienced heightened interest in the research community regardless of the fact that the model was first proposed several decades ago by Curtis and Jackson (1962). These researchers recognized that, in specific cases, measures of the same construct exhibit no, or even negative, correlations. Accordingly, it is assumed that measures affect the latent variable or are its causes and not its effect. Hence, as the indicators represent distinct causes, only their co-occurrence determines the meaning of the latent construct. It follows that formative measures are not interchangeable (Jarvis et al., 2003; Rossiter, 2002) and omitting an indicator potentially alters the nature of the whole construct (Bollen & Lennox, 1991; Diamantopoulos et al., 2008).

Having said this, the measurement model mirrors this research understanding of the focal construct being best measured with reflective indicators on the item- as well as overall construct-level. More specifically, this thesis formally models eWOM trust as a second-order latent construct that has multiple first-order sub-dimensions as reflective indicators and the subdimensions themselves have multiple reflective observable indicators. This implies that it is assumed that observable indicators reflect the specific characteristics of the latent subconstructs. According to Jarvis et al. (2003, p. 202), with such reflective indicators “the direction of causality flows from the construct [i.e., sub-dimensions or types of eWOM trust] to the measures [i.e., observable indicators]”. Further, the conceptualization implies the eWOM trust construct to exist on a higher, more abstract level and that its five dimensions are best viewed as consequences of the focal construct. Accordingly, the various types of trust - which represent different aspects of trust and mirror a common orientation towards the trust object - are considered to reflect a person’s “overall” eW OM trust as their overall latent variable (Bollen & Lennox, 1991; Diamantopoulos et al., 2006). In other words, the first-order latent constructs are a manifestation of the same focal construct (Wong et al., 2008). Hence, eWOM trust as a construct is assumed to exist separately at a deeper and more embedded level than its subdimensions, and further it is assumed that changes in the latent construct (increases/decreases in the person’s global orientation towards OCR) would lead to simultaneous changes in all of its sub-dimensions (MacKenzie et al., 2011) or the person’s trusting beliefs, attitudes, and intentions to rely. It seems appealing to theorize that if individuals adopt their general disposition to trust eWOM, they do this by allowing their general attitude towards the trust object consistently to influence their thoughts, feelings, and conations. A well-trusting person is hence unlikely to have a high belief in the OCR’s honesty while not having the opinion that it is benevolent. In literature, higher-order constructs of this kind have been called totally disaggregated constructs by Bagozzi and Heatherton (1994), superordinate constructs by Edwards (2001) and Williams et al. (2009), or type I constructs by Jarvis et al. (2003).

Such an approach is consistent with earlier conceptualizations of trust and closely related concepts (e.g., Casalo et al., 2011; Corbitt et al., 2003; Flavian & Guinaliu, 2006). However, this research also acknowledges the rising debate concerning appropriate measurement models suitable for trust-related constructs. Such a discussion is justified, as various researchers have misspecified trust-constructs in different contexts. This is because there is more than one type of trust construct.

When conceptualizing eWOM trust, it was understood that constructs are per se not inherently reflective or formative (Baxter, 2009; Diamantopoulos, 2011; Wilcox et al., 2008). In fact, it is particularly the characteristic of the latent variable conceptualized by the researcher and also the content of the indicators which both determine how the construct and its indicators are related and whether the use of reflective or formative indicators is appropriate (MacKenzie et al., 2011, p. 302). According to Hildebrandt and Temme (2006), the underlying research question dictates whether a reflective or formative measurement approach is appropriate to operationalize the construct.

In general, attitudinal constructs can be typically operationalized with both approaches (Albers & Hildebrandt, 2006). While the nature of the observable (reflective) indicators due the items used may not give rise to controversy, the conceptualized causal relationship between the subdimensions (types) of trust and the overall eWOM trust construct may have to be clarified. The proposed measurement model is deducible from prominent literature strings in trust research, as well as personality research. The latter emphasises that trait-like constructs like the disposition to trust exist, which are stable over time and are closely linked to a person’s identity. This literature provides a valuable auxiliary theory (Bagozzi, 1982; Blalock, 1969) that enables this thesis to make a conceptual decision about the linkages between the construct of interest and its sub-dimensions. The previous chapters (and folk wisdom) tell that individuals are able to develop trust in various objects, contexts and situations. Trust can arise suddenly due a calculation process or current feelings about the trust object (which are therefore environmentally caused) (i.e., situational or initial trust), but trust can also exist on the personality level of a person (i.e., generalized trust). This makes it unrealistic to argue for the existence of a single, generalizable trust construct. While the various trust conceptualizations seem to share some common ingredients, it is the close context but also the specific domain which both determine the most appropriate measurement model in the eye of the researcher.

Since the nature of the trust construct is dependent on various circumstances and therefore specific to the research object, it is inadequate to argue for a single measurement model that captures the construct under any circumstances.

For instance, some trust researchers explicitly (or implicitly) advocate the view of what Jarvis et al. (2003) describe as a type II construct, where the sub-dimensions of trust are considered causal for the trust construct. Such a conceptualization is in accordance with a distinct literature stream, understanding trust as consisting of different types or levels of trust (e.g., Boon & Holmes, 1991; Lander et al., 2004; Panteli & Sockalingam, 2005; Zaher et al., 1998). Earlier conceptual contributions have confirmed the existence of subsequent trust levels, as well as the fact that different components of trust become operational at specific trust levels. By adapting the view of Boon and Holmes (1991), Shapiro et al. (1992), as well as Lewicki and Bunker (1995), numerous authors (e.g., Ba, 2001; Boon & Holmes, 1991; Lander et al., 2004; Zaheer et al., 1998) agree that trust is a dynamic process and it possesses a distinct character in different stages of relationships (Hsu et al., 2007). Trust is here best viewed as a phenomenon that is time-consuming to engender, hence, trust is developed through repeated interactions over time. It’s very unlikely that individuals are able to perceive certain degrees of trust without a minimum of previous experience. However, trust is finally situationally constructed, fragile and it can be easily destroyed.

The above-cited scholars furthermore advocate the proposition that trust develops in sequential stages. For example, due to the work of Panteili and Sockalingam (2005), three stages of trust exist which are linked in a sequential iteration. This hierarchical approach to trust means that the achievement of trust at one level enables the development of trust in the next level. Additionally, the trust conceptualization of Sheppard and Sherman (1998), for instance, identifies four distinct and ordered forms of trust: (i) shallow dependence, (ii) shallow interdependence, (iii) deep dependence, and (iv) deep interdependence. In a similar vein, Barney and Hansen (1994) conclude that trust strength can be categorized as weak, semi-strong, and strong forms of trust. Implicitly, McAllister (1995) has contributed to a better understanding of trust levels by proposing two different elements of trust: (i) cognitive-based trust, defined as an individual’s rational use of evidence and analysis to form beliefs and attitude towards a trusted party, can be easily inferred as representing a low level of trust. In contrast, (ii) affect-based trust, which aims to measure the emotional bonds among truster and trustee, signals a high level of trust (Chen & Dhillon, 2003). This view is also supported by Wicks et al. (1999). Lewicki and Bunker (1995) add that trust can be conceptualized in three sequentially ordered and linked forms of trust. (i) The lowest level of trust but also weakest form of trust is calculus-based or cognition-based trust. Here, trust is simply based on the truster’s beliefs about the trustee’s characteristics and thoughts about the trustee’s future intentions and behaviour. Calculus-based trust can be typically found in new or first-time relationships. (ii) In contrast, knowledge-based trust can only be formed over time, where the interaction partners have the opportunity to increase their knowledge about each other. Here, the level of trust is more distinctive, as the parties form relatively strong bonds, needed to establish an exchange relationship. (iii) Identification-based trust mirrors the highest and strongest level of trust, which is least fragile to changing environment. Here, the parties share common interests as well as values. This view is also consistent with the common argument that trust shall be deemed to develop over time and repeated interactions (e.g., Mayer et al., 1995; Rempel et al., 1985; Silence et al., 2006) but can also arise in the first encounter of interaction partners (e.g., Stewart, 2003; Trifts & Haubl, 2003).

Other researchers have supported the view that the individual’s perceptions of the trust types (i.e., trusting beliefs, attitudes, and intentions) can be used to infer levels of trust. For instance, Grimes (1978) as well as Wicks et al. (1999) suggest that the attitude of confidence represents a heightened level of trust. There is additional evidence that trusting intentions, as they involve a person’s willingness to take a risk or to be vulnerable, implies high trust.

The thesis at hand recognizes and values the contributions of researchers who have discussed how trust changes with the passage of time. However, having conceptualized eWOM trust as a relatively stable orientation to rely on online customer reviews and recommendations that has been developed with prior experiences, this thesis does not see that specific types of trust become operational at different stages of trust. Rather, in line with trait research, it proposes that the relationship between the different types of trust (i.e., dimensions of eWOM trust) is not incremental and additive, but consistent. The general nature of the relationship determines its consequences and not the kind of relationship between the trust object and the truster. Accordingly, it is assumed that a person’s high levels of eWOM trust should be mirrored by adequate levels of trusting intentions, trusting attitudes as well as trusting beliefs at the same time. Individuals tend to have a uniform orientation and are always able to express this conformity. Psychological literature is far from answering the question whether cognitions precede emotions or vice versa and controversies long dominate the scientific discussion (Izard, 1991, 1992; Leventhal & Schwerer, 1987). For instance, Zajonc (1984) argues that affect and cognition are separate and partially independent constructs and that affect can be generated without any prior cognitive process. Contrasting with this view, Lazarus (1984) characterizes emotions to be completely intertwined with cognitive processes. Accordingly, the theoretical background of above-cited literature may also be compromised, as the problem concerning the sequence of occurence of the levels or types of trust is also not completely resolved.

In this thesis’ conceptualization of the eWOM generalized trust orientation, the construct is simultaneously mirrored by discrete, but related types and elements of trust. Such a view emphasizes trust as a trait-like social concept leading to cross-situational and cross-personal enduring individual differences (Costa & McCrae, 1995; Murtha et al., 1996). It can be that some situations induce slight variations in the level of trust, but in general it remains stable over time and context (Hogg & Vaughan, 2011). Concerning its conceptual nature, eWOM trust is hence more closely related to an individual’s disposition to trust (Bowlby, 1973; McKnight et al., 1998; Rotter, 1967, 1971) than a situation-dependent construction of trust. Scale development literature regularly demands from multi-dimensional constructs representing a general personality trait that the relationship between the construct and its sub-dimensions are captured by the superordinate construct (MacKenzie et al., 2011; Williams et al., 2009). For instance, for Fornell and Bookstein (1982, p. 292), “constructs such as “personality” or “attitude” are typically viewed as underlying factors that give rise to something that is observed. Their indicators tend to be realized, then as reflective”. This is also acknowledged by Albers and Hildebrandt (2006), who state that a reflective measurement approach is appropriate for most attitudinal/behavioural constructs.

According to MacKenzie et al. (2011), a second-order measurement model with multiple first- order dimensions as reflective indicators is appropriate if (i) the latent construct is stable (i.e., an individual difference variable) over time or across situations; and (ii) the focal construct is measured with randomly selected indicators, each of which is reflective of a focal construct. The latter implies that the sub-dimensions are different manifestations of the second-order factor (MacKenzie et al., 2011). A change in any of the first-order factors is therefore assumed to result in a change in the other underlying indicators (Serva et al., 2005). It follows that first- order constructs have to be strongly correlated to support a reflective conceptualization. Numerous studies emphasizing trust and trust-related constructs have revealed that a high degree of intercorrelations between trust elements (i.e., trusting beliefs) are regularly found (e.g., Bhattacherjee, 2002; Gefen, 2002b; Gefen et al., 2003; Hsio et al., 2010; Mayer & Davis, 1999; Murphy, 2003; Schoorman et al., 1996). Other studies have identified a reasonable relationship between trusting beliefs and trusting attitudes, as well as trusting intentions (e.g., Benedicktus et al., 2010; Chang & Chen, 2008; Chen & Barnes, 2007). Such positive intercorrelations imply the existence of an underlying higher-order factor (Bollen, 1989) - or what is here called eWOM trust.

The above-cited literature provides evidence that the application of a type I construct in this research context is appropriate, as trusters tend to have uniform perceptions of the different aspects of the trustee (here, online reviews/reviewers), personal experiences and behavioural intentions. That is, individuals maintain consistency in their evaluations, especially when these provide the basis for a constant personality trait (Serva et al., 2005). Such a perspective is supported by a number of cognitive consistency theories (Abelson et al., 1968), including Festinger’s (1957) cognitive dissonance theory and Heider’s (1946) balance theory. In their essence, these theories argue that human beings strive to resolve conflicting perceptions - they try to maintain internal consistency and agreement among their various cognitions, beliefs, and feelings (Hogg & Vaughan, 2011). For example, cognitive dissonance theory assumes that if there are inconsistencies between an individual’s attitudes and his/her behaviour, the person will impose changes in order to resolve the currently existing dissonance. One way that individuals can reduce dissonance is by altering the dissonant beliefs so that they are no longer inconsistent (Festinger, 1957). Human beings constantly try to confirm existing perceptions about other persons or objects. This also holds true in situations where individuals have to accept beliefs or attitudes that lack reasonable foundation (Kahnemann et al., 1982; Mynatt et al., 1977), or when they have to reject conflicting information in favor of confirming evidence (Serva et al., 2005). In order to explain such consistency in individuals’ mental processes, type I constructs are regularly used in marketing (e.g., Huang et al., 2011; Riefler et al., 2012) as well as trust research approaches (e.g., Casalo et al., 2011; Wang & Benbasat, 2005). The process described above is analogous to this research’s second-order conceptualization of eWOM trust. Consumers are likely to form consistent perceptions of trusting beliefs, attitudes as well as intentions in order to justify their general level of trust towards and usage of opinions of the reviewers. A consumer who believes in the integrity of online reviews has at the same time a strong willingness to rely on such a source of information - and at the same time will show approach behaviour towards OCR. Attribution theory (Hogg & Vaughan, 2011) or the constant human process of assigning a cause (that is, having favourable thoughts, feelings, and intentions towards the trust object) to our behaviour (i.e., regular eWOM usage) may also help to explain this consistency. This is regarded as a meaningful strategy to avoid dissonance and express inner, consistent orientation.

The thesis further argues that the content of eWOM trust is determined by the general possibility of developing a certain position towards the different aspect of trust, as well as the kind of relationship between the consumer and the trust object. Concerning the former, it has to be mentioned that eWOM trust assesses a person’s general and relatively stable reliance on a specific information source. This reliance has been developed over time and has therefore nothing to do with evaluating a trust object in a specific situation that is maybe new to the consumer. Rather the individual had the time to develop a certain orientation towards the trust object and holds a cognitive, emotional and conative position towards the trust object. The eWOM trust scale hence is intended to capture the true level of trust in situations where the truster has (at least some) prior experience with the trust object. In contrast, trust in the initial stage of a relationship or contextual development of trust in specific OCR (i.e., initial and situational eWOM trust) is likely to be characterized by other parameters. Here, the truster may not be able to form an attitude or develop emotional bonds yet. Therefore, measuring trust with affective trust types such as trusting attitude may not be appropriate at this stage (c.f., Kim & Tadisina, 2007). This approach is also implicitly taken by various scholars who have investigated initial trust. For instance, the research schools around McKnight and Gefen have restricted their operationalization of first-time trust on trusting beliefs which are predominantly cognitive in nature. Similarly, various other trust researchers have concentrated intensively on the truster’s cognitive beliefs when discussing the components of trust. However, a recognizable stream of literature emphasizes that emotional orientations as well as behavioural


intentions have to exist together with a cognitive component in order to represent a construct that can be called trust (Johnson & Grayson, 2005; Lewis & Weigert, 1985).

Another proposition of this literature string which is also picked up in this thesis’ construct conceptualization is that the relative importance of the trust components is dependent on the kind of (social) relationship between the truster and the trustee (Lewis & Weigert, 1985). Accordingly, for primary-group relationships (e.g., friendship), trust can be best described in terms of affective aspects, while cognitive aspects are likely to dominate in the context of secondary-group relations. The dominant role of cognitive trust has also been found in the context of social interactions with secondary groups (Johnson-George & Swap, 1982; Rempel et al., 1985). Basically, when consumers develop trust in online customer reviews and recommendations over time they form a general orientation towards a social institution or a kind of secondary-group relationship. This implies that eWOM trust is mainly based on cognitive trust components.

This research acknowledges that the concept of eWOM trust does not belong to the group of latent constructs where the causal direction between indicators and their respective constructs (e.g., Diamantopoulos & Winklhofer, 2001; Podsakoff et al., 2003) can easily be defined. The true nature of trust is that a generalizable and applicable “trust construct” does not exist. Rather, trust has to be conceptualized in terms of its specific object, context and scope. The role of the “tripiad” may be differential in these various trust contexts. Hence, the focal construct has to be investigated from a larger research context, and the appropriateness of its measurement by means of several criteria (Wilcox et al., 2008). In such a complex conceptual environment, it may be sometimes more appropriate to measure trust with an index (Diamantopoulos & Winklhofer, 2001), while in other cases the development of a trust scale based on conventional scale development procedures (e.g., Churchill, 1979; DeVellis, 2012) may be more constructive, as it was considered in this research. This thesis argues that viewing trust as what Jarvis et al. (2003) called a type II construct may be appropriate in specific trust contexts only. More specifically, it is assumed that a formative view of trust on the second-order level is adequate when trust is understood as a situational variable which results from a contextual judgment process of specific characteristics of the trust object (the reviews just read), such as trusting beliefs (e.g., situational eWOM trust). Here, the elements of trust are likely to be (more or less) evaluated independently from each other and together “form” a latent construct which represents overall trust. In such a research context, trust is typically conceptualized as a situation-specific variable towards a particular trust object (e.g., Jarvenpaa et al., 2000; Kim et al., 2008) and consumers are regarded as being able to make independent judgments on the discrete elements in order to make an objective trust decision. When a consumer faces a specific review, he/she will decide upon (mainly cognitive) judgments about the characteristics of these reviews as well as upon personal feelings and general orientations whether or not he/she has trust in the opinions of others. (However, this judgment would be heavily influenced by generalized eWOM trust, as outlined before.)

Some researcher state that decisions about the nature of the measurement model should not be taken lightly, as assuming wrongly a reflective measurement model where a formative would be appropriate leads to expansive errors in model specification - or what some researchers call a type I error (Diamantopoulos & Siguaw, 2006; Jarvis et al., 2003; Petter et al., 2007). Various studies attempted to gain insight into the consequences of model misspecification. For instance, by reviewing an earlier study (Cadogan et al., 1999), Diamantopoulos and Siguaw (2002) demonstrate that different measurement models lead to different sets of indicators. This variation finally results in a completely different meaning of the constructs involved. Albers and Hildebrandt (2006) relax this insight and state that specifying a measurement model “wrongly” only results in a restricted model and not an erroneous one per se. Differences in results remain small as long as one refrains from excluding specific items. This said, the eWOM trust scale developed in this research is based on a literature-based construct conceptualization deemed to be appropriate for a stable, trait-like construct. Assuming that the five dimensions are different manifestations of the same construct (Wong et al., 2008), and assuming that “eliminating any of them would restrict the conceptual domain of the construct” (MacKenzie et al., 2011, p. 301), this thesis is confident that formally modelling eWOM trust as a (reflective) second-order construct with five (reflective) first-order dimensions is appropriate.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science