Toward Understanding Trust: A Response to Adams and Miskell

Abstract

Purpose: Trust is a key component of successful schools. Although scholars widely agree that trust is multifaceted, there is less agreement about the number and nature of these factors. In the October 2016 issue of Educational Administration Quarterly, C. M. Adams and Miskell argued that their Teacher Trust of District Administration Scale provided evidence that trust is a single factor that cannot be unpacked, and that our three-factor theory of trust, which asserts that trust involves the discernment of benevolence, competence, and integrity, is invalid. We find multiple conceptual and methodological flaws in their reasoning. Method/Approach: We analyze data provided by C. M. Adams and Miskell that were used in their original 2016 article. The data set includes responses from 606 teachers in 72 schools to the 10 survey questions that comprise the Teacher Trust of District Administration Scale. We reproduce and critique the results of four models presented, and corrected, by C. M. Adams and Miskell, and present an alternative second-order model of trust with three first-order factors representing benevolence, competence, and integrity. Findings and Implications: Consistent with theory, we find that trust is more appropriately modeled as a multifactor construct. A multifactor model of trust is not merely an advance in measurement, it has important, actionable implications for research and practice. Measuring trust as a second-order factor, with first-order factors benevolence, competence, and integrity, positions us to make more nuanced judgements about trust, more easily diagnose problems, and prescribe interventions needed to develop, maintain, or repair trust in schools.

Keywords

trust teacher trust district trust student trust school improvement schools

In the October 2016 issue of Educational Administration Quarterly (EAQ), C. M. Adams and Miskell published an article titled “Teacher Trust in District Administration.” Our close attention was drawn to this article because it explicitly challenged the findings of two publications in which we had independently sought to describe the factor structure of trust as a component in successful education programs (Makiewicz & Mitchell, 2014; Romero, 2010). We examined the published data and analyses initially to determine whether its critique of our work would require us to acknowledge that our prior work was flawed. On close reading, however, we found problems in the published article at three levels. First, while the published article presented a correlation matrix for the data used in the study, we were unable, using that matrix, to reproduce the published model parameters and fit statistics. Second, even after correcting the reported statistics, we found the methods used to be inadequate to support the conclusions drawn from the data. And third, we disagreed with the notion that a model of trust could consist of five facets that, as the article concludes, were independent measures of trust components that cohere in a single factor whose constituent subfactors cannot be modelled. We take up these issues in order below.

Before turning to discussion of these issues, however, we want to acknowledge that the article’s senior author, Curt Adams, has been most cooperative in sharing the original data for this study and clarifying publication errors. We and Curt Adams are in complete agreement that all of the model statistics reported in his letter to the EAQ editor and our presentation in this article are accurate—the models have been tested with both the IBM/SPSS AMOS© structural equation package and the MPlus© statistical modeling program. We tell the story of investigating the errors in the original article because we see this story as a helpful way of critically digesting published research and showing how our thinking developed and our conclusions were reached.

Reporting Errors

Our first step in considering whether the C. M. Adams and Miskell (2016) article would require us to abandon our previously published work was to reproduce the analyses published in their article. The article contains a correlation matrix, means, and standard deviations for the study data (p. 14 in C. M. Adams & Miskell, 2016), which we copied into both AMOS and MPlus data formats and ran the models described in the published article. To our surprise, neither the model fit statistics nor the published regression coefficients matched our outputs. Initially, we thought that the problem might lie with the fact that the correlation matrix in the published article was limited to two significant digits, so we contacted the senior author to solicit the original data set so as not to be limited by potential round-off errors.

Subsequent analysis of the full data set revealed that our inability to replicate the results was not due to rounding errors but due to multiple errors in the analysis and reporting, starting with the measurement model. These mistakes are addressed by the authors in their letter to the editor, and we do not repeat them here. However, even with these errors corrected, significant issues remain with the analysis, and most important, with inferences that can be drawn about the nature of trust. It is these issues that are the focus of our response.

Toward Understanding Trust

What is the factor structure of trust? In “Teacher Trust of District Administration: A Promising Line of Inquiry,” C. M. Adams and Miskell (2016) argue that trust is a single-factor, one-dimensional construct that cannot be unpacked. They assert that an alternative three-factor theory of trust, put forth by Makiewicz and Mitchell (2014) and Romero (2010), lacks validity. However, this ambitious claim is conceptually and methodologically flawed. Even after computational errors are corrected, the models presented in the C. M. Adams and Miskell (2016) article demonstrate poor fit, are not aligned with theory, are driven by measurement ambiguity, and make claims that reach beyond the limitations of the data. In declaring that they have developed a reliable instrument for measuring district trust, unsubstantiated conceptual and logical leaps were made. To be clear, we are not referring to errors that the authors have recognized and corrected in the letter “Teacher Trust in District Administration: Correcting the Evidence” (C. Adams, 2017). Relying on their reanalysis of data published in this issue of EAQ, and on some additional analyses we present below, we are focused on more interesting and important issues. The issues we are raising have compelling implications for future directions in research and theory about trust.

Trust in Schools

There is a substantial and growing body of research documenting the importance of trust in high functioning, modern institutions (Putnam, 2001; Fukuyama, 1995; Kramer, 1999; Schoorman, Mayer & Davis, 2007), among them schools (Bryk & Schneider, 2002; Tschannen-Moran, 2004; Forsyth, Adams, & Hoy 2011; VanMaele, Forsyth, & Van Houtte, 2014). Mitchell (2015) recognizes trust as the defining characteristic of professional work, distinguishing it from the performance of labor, craft, or artistic work activities. The literature on the role of trust in schools is extensive, and we do not attempt a comprehensive review here. Briefly, however, research indicates that teacher and principal trust play a role in efforts to collaborate, and in their openness to innovation, mentoring, and professionalism (Celano & Mitchell, 2014; Cosner, 2009; Fisler & Firestone, 2006; Gray, Kruse, & Tarter, 2015; Hallam, Dulaney, Hite, & Smith, 2014; Hoy, Gage, & Tarter, 2006; Makiewicz, & Mitchell, 2014; Seashore-Louis, 2007; Tschannen-Moran, 2009; Tschannen-Moran & Gareis, 2015). Student trust is associated with academic optimism; belonging; identification with school, behavior, discipline, academic achievement; and college ambition (Gregory & Ripski, 2008; Romero, 2015; Schneider, Judy, Ebmeye, & Broda, 2014; Tschannen-Moran, Bankole, Mitchell, & Moore, 2013). Trust has been shown to be key to effective partnerships with parents and families (C. M. Adams, Forsyth, & Mitchell, 2009; K. S. Adams & Christenson, 2000; Forsyth et al., 2011; Knopf & Swick, 2007). Of late, researchers have begun to investigate the role of trust with the district office. Daly and Finnigan (2012) examined principal trust of the district office, and most recently, C. M. Adams and Miskell (2016) set out to investigate teacher trust of the district office.

That trust is essential seems clear. But how do we develop, maintain, or repair trust in schools? To answer this question, we need a clear understanding of what trust is, and the means to measure it reliably. If trust is low, to borrow a metaphor from medicine, we need to be able to diagnose the problem and prescribe effective remedies. Unfortunately, as the overall corpus of the scholarly literature on trust makes clear, trust is a latent construct. It involves complex relationships that are difficult to quantify. Latent constructs, such as motivation, self-determination, self-efficacy, and numerous other concepts in social science, are those we know exist but for which there is no single direct measure. Latent constructs are measured indirectly via subdimensional indicators that reflect the integrative construct. Discovering good proxy measures is hard work and must begin with theory. That is, we must begin by asking what trust is conceptually, not just as a statistical artifact.

While there is wide agreement, across and beyond the field of education, that trust has multiple dimensions, there is less agreement about the nature and number of these qualities (Romero, 2010). A number of attributes have been suggested. Among them, benevolence, competence or ability, integrity, regard for others, respect, reliability, openness, fairness, and honesty are most often cited. Mayer, Davis, and Schoorman (1995) identified three components of trust: benevolence, ability, and integrity. Mishra (1996) posited competence, concern, reliability, and openness. Hoy and Tschannen-Moran (1999) pointed to five attributes: competence, benevolence, reliability, honesty, and openness. Bryk and Schneider (2002) identified the following: competence, regard for others, respect, and integrity. Other research suggests that trust may have as many as 10 subcomponents (Butler, 1991).

While it is possible that more than three constructs make substantial contributions to the development of trust in organizational relationships, we agree with Mayer et al. (1995) that a “parsimonious model . . . with a manageable number of factors should provide a solid foundation for the empirical study of trust” (p. 711). Our work, building on that of others, demonstrates that there are at least three distinct conceptual components to organizational trust. We hypothesize that benevolence, competence, and integrity best describe these three conceptually distinct components of trust. While it may eventually be shown that there are other factors that need to be taken into consideration, or that one or more of these three basic factors should be differentiated into two or more independent factors, we are confident that there are at least three interrelated but substantially independent factors to be considered. Theorists may develop fine lines of distinction identifying more factors, but arm chair discussions separating, for example, “regard for others” from “respect,” or “openness” from “honesty,” are not yet yielding measures reliable enough to be discerned empirically. That is, words whose meanings overlap in a dictionary or a thesaurus run a good chance of being empirically indistinguishable or multi-collinear because they share attributes of the same basic or highly related constructs.

In Table 1, we have summarized what appear to be the close parallels among similar conceptual terms used in well-regarded pieces of research. The major terms are organized under the three headings we believe best represent the principal components of trust. As shown in the table, competence (or ability) is universally regarded as a facet of trust. It is arguable that “regard for others,” “concern,” and “respect” all fit under the umbrella of benevolence. Fairness, reliability, honesty, and openness may be appropriately thought of as relating to integrity and, in fact, are synonyms in Roget’s Thesaurus (Kipfer, 2001).

Table 1.

Dimensions of Trust.

Study	Competence	Benevolence	Integrity
Mayer et al. (1995)	Ability	Benevolence	Integrity
Mishra (1996)	Competent	Concern	Reliable; Openness
Bryk and Schneider (2002)	Competence	Regard for others; Respect	Integrity
Hoy and Tschannen-Moran (1999)	Competence	Benevolence	Reliability
Tschannen-Moran (2004)			Honesty
Forsyth et al. (2011)			Openness

In short, the literature makes clear that trust is composed of multiple, interrelated, yet conceptually distinct constructs. As described in the next section, and in contrast with C. M. Adams and Miskell (2016), our work takes this into account, explicitly modeling trust to reflect this. In doing so, we make clear that this distinction transcends measurement; it has implications not only for future research but also for actionable practice (Schneider et al., 2014).

Trust: Benevolence, Competence, and Integrity

In our published work, we draw from Mayer et al. (1995) and hypothesize that trust consists primarily of the three factors organizing Table 1: benevolence, competence, and integrity (Makiewicz & Mitchell, 2014; Romero, 2015). This model is not only parsimonious and intuitively appealing, but as Table 1 indicates, it is also aligned with extant theory that trust is a complex meta-construct with multiple constituent latent constructs. Benevolence is the sense that the trusted party has the trustee’s best interests at heart; competence reflects the belief that the trustee has needed skills and abilities; and integrity reflects the belief that the trustee will behave fairly and ethically. Though the decision to trust involves the perception of all three of these elements, each is conceptually distinct. Importantly, we measure trust to reflect this by modeling it as second-order factor with first-order factors benevolence, competence, and integrity. A second-order model is appropriate, not only because it is driven by theory rather than by convenience of measurement, but notably because it enables us to more clearly probe the components of trust (Chen, Sousa, & West, 2005; Koufteros, Babbar & Kaighobadi, 2009; Rindskopf & Rose, 1988).

Trust is modeled as a second-order factor drawing on the shared common variance of the first-order factors comprising benevolence, competence, and integrity. The separate first-order factors are free to vary independently. Therefore, although trust requires all three, they can be present in varying proportions. A low level of trust may be recognized in relatively equal low scores on all three components. Or, low trust may be granted to a trustee who is characterized by high levels of one component, say competence, while lacking in one or both of the others. The utility of such a model is that, where trust is low, it is possible to more precisely identify the cause and develop a remedy. For example, a new superintendent may not be trusted by teachers because they believe he lacks the skill set necessary to run the district (competence). In this case, developing trust might require the superintendent put together a solid budget, get staff hired on time, and keep facilities in good repair. It is also possible that teachers may perceive a new superintendent to have the skill set necessary to run the district, but not trust that he is benevolent because he embraces charters or has a history of being combative with teacher unions. In this case, remedial action might focus on relationship building.

Five Facets

In contrast to our model, C. M. Adams and Miskell (2016) hypothesized that trust consists of five components—benevolence, competence, honesty, openness, and reliability. After carefully describing the components, and spending time explaining the differences among them, however, they argue that these concepts constitute not “factors” but “facets.” By this, they mean that trust is a unitary construct composed of a single conceptual factor. Using the 10-item “Teacher Trust in District Administration” survey published in the same article, they try to present evidence supporting this conclusion. To parse their evidence, they used the SPSS/AMOS© structural equation modeling software to develop four-factor analysis models based on 606 teachers located in 72 schools in a single school district. Their survey asked teachers to respond to 10 statements about district administrators. For example, teachers were asked if district administrators valued the expertise of teachers, showed concern for the needs of their school, followed through on commitments, and had a coherent strategic plan. Taking a stab at testing for the existence of five independent factors, C. M. Adams and Miskell (2016) used two questions written to assess each of the five hypothesized facets of trust (benevolence, competence, honesty, openness, and reliability).

In their first “original hypothesized” model, C. M. Adams and Miskell (2016) model trust as a single factor, measured by all 10 variables. As originally published, this model appeared to be a good fit to the data. In conversation with Curt Adams, however, it was learned that the goodness of fit reached an acceptable level only by including 10 unreported correlated error terms. When rerun without correlated error terms, fit statistics (documented in the C. M. Adams letter of correction) reveal that this model is actually a poor fit with a chi-squared of 442.73 (df = 35), χ²/df ratio of greater than 12, root mean square error of approximation (RMSEA) of .14, comparative fit index (CFI) of .93, and Tucker-Lewis index (TLI) of .91. Rather than being accepted as evidence of a single factor with acceptable construct validity, this model should have been rejected.

C. M. Adams and Miskell (2016) then presented a second, “trimmed” model that examines the factor structure of 5 of the 10 original trust indicators. Unfortunately, C. M. Adams and Miskell use two questionable criteria in trimming five of the indicator variables from the original model. First, while hypothesizing that trust is a single-factor construct, they, nevertheless, insist on retaining for the trimmed model one indicator from each of their hypothesized “facets,” implicitly embracing a multifactor conceptualization (this point we will return to below). The second questionable procedure was to capitalize on chance by retaining the highest loading variable for each facet. In dropping the less well correlated variables they virtually assure a superior fit for the trimmed model, but the better fit is due to data selection rather than improved theoretical analysis. Moreover, it is important to note that the highest loading variables for each of the five facets were not the five highest loading variables in the original model. Selecting the five highest loading variables would have meant the trimmed model would not include all of their five facets. As in the first model, the article failed to report five correlated error terms that improved the fit of the model. Described in the C. Adams (2017) letter, the trimmed model was rerun. Two changes were made in rerunning this model. First, due to a mix-up in variable labeling the trimmed model did not have one variable from each of the five hypothesized facets (and thereby presented erroneous results). Second, the model was rerun without the (unreported) correlated error terms. Results for the corrected model should also be rejected as providing evidence of a single factor with five independent facets. The corrected model produced a chi-squared of 87.39, which is smaller than that for the original model, but this is largely a byproduct of the reduction in the number of degrees of freedom (from 35 to 5). The CFI improves to .97, and the TLI is a modest .94, but other indicators are more problematic. The ratio between the χ² and df is too high at 17.48, and the RMSEA increases to .17, well beyond the generally accepted indication of a good fit. In both the original and trimmed single-factor models, the RMSEA points to the significant correlation among the residual terms, which is a clear indication that unmodeled structure remains in the data. That is, the model is underspecified, suggesting the presence of one or more other factors. We return to this below.

C. M. Adams and Miskell (2016) compare their single-factor models to two multifactor models. In their third model, trust is modeled as a second-order factor with the five first-order factors representing their hypothesized five facets of trust: benevolence, competence, openness, honesty, and reliability. This model retains the original 10 variables with 2 measures for each factor. With a chi-squared value of 384.61 (df = 30), RMSEA of .14, and none of the fit indices reaching .95 or above, this model is not a good fit to the data. Additionally, in this model there are three standardized coefficients that exceed the theoretical limit of 1.00. These indications of an ill-fitting model are the result of significant correlations among several residual error terms, making it clear that if there is a factor structure accounting for these data, it is not the hypothesized five-factor structure. Beyond the ill-fitting statistics, however, the hypothesized model is weak because it seeks to define the first-order factors with two measurement variables each. Although the second-order factor makes this mathematically possible, for reasons elaborated below, it is not the best approach to developing measurement models.

In the fourth model, trust is also modeled as a second-order factor with three first-order factors that C. M. Adams and Miskell (2016) hypothesize represent the three factors found in our previous work: benevolence, competence, and integrity. This model retains the original 10 variables thought to define 5 facets of trust. Without specific explanation, the variables are allocated to the three factors in our earlier work in an attempt to show that our three-factor structure, developed from a different set of survey items, is not acceptable. Three measures are used to indicate benevolence, three to indicate competence, and four to indicate integrity. The items originally written to capture openness are allotted to benevolence and competence, and the items written to measure honesty and reliability are allotted to integrity. It is this model that they assert accurately represents the model of trust offered by Makiewicz and Mitchell (2014) and Romero (2010). Though it is the best fitting of the five models tested, this model, like the five-factor model above, has an RMSEA that is unacceptably high (.12). More important, this model exhibits an unacceptable negative residual and one factor loading above 1.0. Because the solution is technically improper, we do not consider the other fit statistics.

To summarize this review of the C. M. Adams and Miskell (2016) analysis, we conclude that the evidence presented in their original 2016 article, and corrected in their letter in this issue, does not support the hypothesized explanations that they have investigated. It is not methodologically appropriate to consider the data to constitute a single factor of either 5 or 10 items. It is not appropriate to consider that the data support either of the multilevel models tested. And, it is not appropriate to consider that they have located or measured a reliable model of trust consisting of five independently measured “facets.”

Analytical and Theoretical Considerations

In addition to the fact that the survey data presented by C. M. Adams and Miskell (2016) do not adequately fit the four models they tested, the approach presented in their article has several analytical and theoretical problems. These problems include the following: (a) lack of alignment between measurement and theory, (b) too few indicators to properly support the complexity of their trust model, (c) ambiguous item indicators, and (d) conclusions not supported by the data.

First, simply put, their claim of a five-facet/single-factor model is not aligned with theory. Scholars consistently agree that trust is multidimensional, a concept that necessarily implies a multifactor model. If, as C. M. Adams and Miskell (2016) assert, trust consists of five factors, or as we assert, trust consists of three factors, a properly constructed model would inform our choice. If results from repeated studies do not reveal a five-factor structure, and they do not, then it is time to rethink theory or reexamine measurement. Claiming instead that these are facets rather than factors side steps more important questions about the nature of trust.

Second, when developing a reliable instrument for measuring a latent construct, it is essential that researchers gather data on enough measurement variables to permit reliable testing of the hypothesized model. A two-step approach is best used in model development (Anderson & Gerbing, 1988). In the case of a second-order factor model, this would mean that measurement models for the first-order factors should be tested and validated as a first step. Once valid first-order factors are created, these factors can then be posited as indicators of a higher, second-order factor (Chen et al., 2005; Koufteros et al., 2009; Rindskopf & Rose, 1988). At an absolute minimum one needs three measures for every independent factor. Importantly, however, a factor with only three items is a just identified model, with zero degrees of freedom, making it impossible to test model fit. Without imposing additional constraints such as equality of regression paths, at least four items are necessary. Thus, hypothesizing a five-factor model implies a survey of at least 20 well-constructed measurement items and more would be better (Marsh, Hau, Balla, & Grayson, 1998). Of course, once a reliable model has been validated, it would be feasible to propose a five-item quick survey provided that the short-form survey is shown to reliably map the trust construct in the same way as an established long-form survey. C. M. Adams and Miskell did not conduct, as is customary, a validation of a full-scale model and have asserted, instead, that the failure of their 10-item survey to document the existence of their five factors provides evidence that these factors don’t exist.

Third, although the authors intended to develop a scale to assess teacher trust in district administration, their measures lack sufficient precision. Survey questions written for the Teacher Trust in District Administration Scale were intended to gauge specific facets of trust. However, when C. M. Adams and Miskell (2016) submitted their questions to a group of educators for feedback on construct validity, those educators identified some of the items as tapping into more than one, or different, constructs. For example, the item District administrators often say one thing and do another was alternatively identified as honesty, openness, and reliability. The item District administrators value the expertise of teachers was identified by most as benevolence, though some felt it reflected competence. The models presented by the authors share the same ambiguity. For example, they stated that the item District administrators are open to ideas about school improvement is intended to measure openness, but then assigned it to benevolence in the three-factor model. Similarly, the item District administrators are transparent in making decisions about district performance, written to measure openness, is later said to tap competence. And items written to measure reliability are renamed integrity. While reliability and integrity are arguably similar, they are not the same. It is quite possible that one can be counted on to reliably act without integrity. This does not mean that the items do not collectively tap into a generalized measure of trust, but it does mean that the Teacher Trust in District Administration Scale is not refined enough to differentiate or delineate factors. In other words, the measures are not unidimensional.

Ordinarily in scale development, when factor analysis reveals that an item taps into multiple factors, and is therefore not unidimensional, that item is dropped in favor of other items that are unidimensional (Marsh et al., 1998). Unfortunately, because their data set is limited to just two questions per facet, C. M. Adams and Miskell’s (2016) ability to drop items is limited. That they are unable to find a factor structure that aligns with theory is not surprising. Valid second-order factor models require valid first-order factor models, and the authors do not have a sufficient number of clear item indicators to adequately measure their first-order constructs (Anderson & Gerbing, 1988; Marsh et al., 1998; Rindskopf & Rose, 1988).

Fourth, perhaps because their composite measure does not facilitate clear examination of key subcomponents of trust, they claim that the components of trust are of equal importance and cannot not be weighed. Yet there is no theoretical reason why important aspects of trust should always be of equal weight, rather than varying based on the nature of the trust relationship, the circumstance, and the setting (Romero, 2010, 2015). As we have already said, if trust is the result of multiple factors working in various combinations, then building or repairing trust may well require very different interventions in different circumstances and settings (see, e.g., Kim, Ferrin, Cooper, & Dirks, 2004; Kochanek, 2005). Moreover, building a bloated factor with many indicators capable of independent influence on the overall factor value does not provide much guidance on how trust may be built or restored. A single-factor model does not easily point to the specific sources of mistrust, and it does not provide guidance on the steps to take to build or restore trust.

In short, there are a number of significant methodological and substantive problems in the original article, and the authors make claims that extend far beyond the limitations of their data. While it is possible that they have tapped into a general, oblique measure of trust, which could be used to draw general inferences about teacher trust of the district office, as C. M. Adams and Miskell do in the second half of their article, the data and analysis is wholly inadequate to conclude that trust is a single factor with five facets, or that our more nuanced three-factor model of trust lacks validity.

A Three-Factor Model of the Data

As we discussed above, the poor fit statistics, in particular the RMSEA, and presence of significant correlated residual terms in the one factor models suggest that there is additional structure in the data. We conducted an exploratory analysis of C. M. Adams and Miskell’s (2016) data set to investigate this possibility. We undertook this with the expectation that since the 10-item survey of teacher trust developed by C. M. Adams and Miskell was intended to define a five-factor or five-facet single-factor model, its items would not neatly fit with our three-factor model grounded in a different theoretical conception of how trust is generated and how it might be repaired where it is lacking. Some of the C. M. Adams and Miskell items, no doubt, lack unidimensional assessment of the three hypothesized first-order latent constructs (benevolence, integrity, and competence) in our theoretical conception. As a result, we expected that some survey items would load on more than one factor. This makes interpretation challenging and means that these items do not define the most useful instrument for assessing teacher trust in district administration. Thus, while these data are problematic if the goal is development of highly reliable measurement scales, our interest is in the factor structure of trust rather than the instrumental reliability of the 10 items. We do not seek, or expect, to find a precise scale, and our goal is data exploration for the purpose of theory building. We ask, given the face validity of the 10 items available to us, “Is there evidence of a three-factor structure in the data, with a meta-factor assessing overall trustworthiness?”

We used a model-generating approach to investigation. In a model-generating approach, an initial theory-based model is specified, tested, and evaluated for goodness of fit. Fit statistics, factor loadings, correlations, residuals, and modification indices are examined and used to make small changes in order to make substantive and statistical improvements to the model (Anderson & Gerbing, 1988). If adequate fit is found, the results from the final model are presented. If no adequate fit is found, we must conclude that either trust lacks the hypothesized factor structure, as C. M. Adams and Miskell (2016) argued and that our three-factor structure should be reconsidered, or the available data are too far removed from the theoretical model being pursued.

Using the data provided by C. M. Adams and Miskell (2016; n = 606), relying on our prior work—the works critiqued in their 2016 article—we modeled trust as a second-order factor, with three first-order factors representing benevolence, competence, and integrity (Makiewicz & Mitchell, 2014; Romero, 2010). Unlike the model tested by C. M. Adams and Miskell (2016), we did not hesitate to allow one or more survey items to be influenced by more than one of the first-order latent factors. Additionally, path coefficients for competence and integrity were constrained to equal in order to generate fit statistics.¹ We insisted that our final model meet two basic benchmarks. First, the tested models had to meet widely accepted statistical criteria for reliability (χ² < 5 times the model degrees of freedom, RMSEA < .08, CFI > .95). Second, the model must reach these fit statistical criteria without modeling any correlated residuals among either the original 10 survey items or the three first-order latent factors. We were able to find and fit a statistically reliable model fitting these two criteria. The text and labels for the C. M. Adams and Miskell (2016) variables are shown in Table 2 (abbreviated labels are reproduced in the final model presented in Figure 1).

Table 2.

Variable Names, Survey Item, and Factor.

Variable	Survey Item (Factor)
FTDist1	Value expertise of teachers (I)
FTDist2	Show concern for the needs of school (B)
FTDist3	Open to teacher ideas about school improvement (B)
FTDist4	Transparent in making decisions about district performance (I)
FTDist5	Say one thing and do another (B, I)
FTDist6	Follow through on commitments (I)
FTDist7	Committed to the stated goals of the district (I, C)
FTDist8	Demonstrate knowledge of teaching and learning (C)
FTDist9	Have established a coherent strategic plan (C)
FTDist10	Take personal responsibility for their actions and decisions (C)

Figure 1.

Model diagram with standardized path coefficients.

As shown in Figure 1, the acceptably fitted model finds three variables loaded on benevolence, five loaded on integrity, and four on competence. As expected, two survey items (Items 5 and 7) are each cross-loaded on two of the first-order factors. Item 5 is predicted by both benevolence and integrity. Item 7 is predicted by both integrity and competence. From the Item 5 cross-loading we can infer that both the district administrators having either a more benevolent attitude, or being recognized for higher integrity, are likely to be seen as NOT being ready to “say one thing and do another” (this item was reverse scored). From the Item 7 cross-loading we can infer that district administrators seen as either more competent, or having higher integrity, are more likely to be seen as being “committed to the stated goals of the district.”

Construct reliability for the tested factor structure was assessed using Raykov’s reliability coefficient (ρ) and Cronbach’s alpha and is reported in Table 3; each of the latent factors demonstrated construct validity with all values well above .70.²

Table 3.

Reliability Coefficients.

Factor	Raykov’s ρ	Cronbach’s α
Benevolence	.79	.91
Competence	.86	.93
Integrity	.86	.93

Overall fit statistics with no correlations among any residual terms show that the model is a reasonably good fit with χ² = 121.19 (df = 31), χ²/df ratio of 3.91, RMSEA of .07, CFI of .99, and TLI of .98. All of the path coefficients are statistically significant (p ≤ .05). Thus, we have a model with three first-order factors and one meta-factor that meets our criteria of reliability and lack of substantial correlations among residuals.

Path coefficients linking the first-order factors of benevolence, competence, and integrity to the meta-factor attributable to trust are powerful and reliable. Approximately 76% of the variance in benevolence (r² = .76), 90% of the variance in competence (r² = .90), and 94% of the variance in integrity (r² = .94) is common to the trust meta-factor.

Conclusion

We agree with C. M. Adams and Miskell (2016) that “validity exists to the degree that the measure represents the underlying theoretical construct and informs credible judgements about the phenomenon of interest (Cronbach, 1971; Messick, 1995)” (p. 683). Indeed, validity is more important than reliability because an imprecise measure of a correct construct is better than a reliable measure of something that is wrong (Marcoulides & Heck, 1993).

Mistakes aside, C. M. Adams and Miskell’s (2016) analysis seems to exhibit confirmation bias, an inclination to see in the analysis what was believed before the analysis was undertaken, and to ignore the countervailing evidence presented when the analysis is completed. While we were initially troubled by the dismissal of our prior research, that concern prompted us to undertake a thorough review of the available evidence and to recognize some important tensions between theoretical development of valid constructs and analytical testing of how well these constructs fit specific data sets. The bottom line for this review is our confidence that data collected to measure organizational trust parameters are revealing that trust is not best modeled as a single factor structure. If trust has “facets” (because the existence of measured factors representing each facet have not been confirmed) we remain confident that this should be seen as the beginning of instrument development, making the search for consistent and reliable factor structures a high priority for future work. If trust is as important as the literature suggests, we ought to be interested in learning more about it and less concerned with preserving a favorite paradigm.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Lisa S. Romero is an assistant professor in the Doctorate of Educational Leadership Program in the College of Education at California State University Sacramento. Her publications have appeared in the Journal of Educational Administration, Educational Administration Quarterly, and Teacher’s College Record. Her research interests include school improvement, trust, educational politics and policy, equity, and social justice.

Douglas E. Mitchell is a professor of the Graduate Division in the Graduate School of Education at the University of California, Riverside. He is past-president of both the Politics of Education Association and the Sociology of Education Association and recipient of the Stephen K. Bailey Award for research contributions to the Politics of Education. Education politics and policy have been the focus of his research and scholarly writing. Published writings include more than 100 books, journal articles, and monographs covering topics in social science theory and utilization, school program and policy analysis, state legislative decision making, labor relations, teacher incentive systems, public support for public schools, and school board elections. As founding director of the California Educational Research Cooperative, he spent 12 years coordinating the work of a major research center supported by substantial contributions from more than 20 local school districts, county offices of education, and University of California, Riverside. He has extensive experience designing and conducting objective, third-party evaluations for educational programs and policies.

References

Adams

C. M.

(2017). Teacher trust in district administration: Correcting the evidence. Education Administration Quarterly. Advance online publication. doi:10.1177/0013161X17721608.

Adams

C. M.

Forsyth

P. B.

Mitchell

R. M.

(2009). The formation of parent-school trust: A multilevel analysis. Educational Administration Quarterly, 45, 4-33.

Adams

C. M.

Miskell

R. C.

(2016). Teacher trust in district administration: A promising line of inquiry. Educational Administration Quarterly, 52, 675-706.

Adams

K. S.

Christenson

S. L.

(2000). Trust and the family–school relationship examination of parent–teacher differences in elementary and secondary grades. Journal of School Psychology, 38, 477-497.

Anderson

J. C.

Gerbing

D. W.

(1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103, 411-423.

Bryk

A. S.

Schneider

B. L.

(2002). Trust in schools: A core resource for improvement. New York, NY: Russell Sage Foundation.

Butler

J. K.

(1991). Toward understanding and measuring conditions of trust: Evolution of a trust inventory. Journal of Management, 17, 643-663.

Celano

S. M.

Mitchell

R. M.

(2014). Mentoring, trust, and teacher efficacy: A powerful force for new teacher induction. In VanMaele

Forsyth

Van Houtte

(Eds.), Trust and school life (pp. 189-204). Dordrecht, Netherlands: Springer.

Chen

F. F.

Sousa

K. H.

West

S. G.

(2005). Teacher’s corner: Testing measurement invariance of second-order factor models. Structural Equation Modeling, 12, 471-492.

10.

Cosner

(2009). Building organizational capacity through trust. Educational Administration Quarterly, 45, 248-291.

11.

Daly

A. J.

Finnigan

K. S.

(2012). Exploring the space between: Social networks, trust, and urban school district leaders. Journal of School Leadership, 22, 493-530.

12.

Fisler

J. L.

Firestone

W. A.

(2006). Teacher learning in a school-university partnership: Exploring the role of social trust and teaching efficacy beliefs. Teachers College Record, 108, 1155-1185.

13.

Forsyth

P. B.

Adams

C. M.

Hoy

W. K.

(2011). Collective trust: Why schools can’t improve without it. New York, NY: Teachers College Press.

14.

Fukuyama

(1995). Trust: The social virtues and the creation of prosperity. New York, NY: Free Press.

15.

Gray

Kruse

Tarter

C. J.

(2015). Enabling school structures, collegial trust and academic emphasis: Antecedents of professional learning communities. Educational Management Administration & Leadership, 44, 875-891.

16.

Gregory

Ripski

M. B.

(2008). Adolescent trust in teachers: Implications for behavior in the high school classroom. School Psychology Review, 37, 337-353.

17.

Hallam

P. R.

Dulaney

S. K.

Hite

J. M.

Smith

H. R.

(2014). Trust at ground zero: Trust and collaboration within the professional learning community. In VanMaele

Forsyth

Van Houtte

(Eds.), Trust and school life: The role of trust for learning, teaching, leading, and bridging (pp. 145-170). Dordrecht, Netherlands: Springer.

18.

Hoy

W. K.

Gage

C. Q.

Tarter

C. J.

(2006). School mindfulness and faculty trust: Necessary conditions for each other? Educational Administration Quarterly, 42, 236-255.

19.

Hoy

W. K.

Tschannen-Moran

(1999). Five faces of trust: An empirical confirmation in urban elementary schools. Journal of School Leadership, 9, 184-208.

20.

Kim

P. H.

Ferrin

D. L.

Cooper

C. D.

Dirks

K. T.

(2004). Removing the shadow of suspicion: The effects of apology versus denial for repairing competence-versus integrity-based trust violations. Journal of Applied Psychology, 89, 104-118.

21.

Kipfer

B. A.

(Ed.). (2001). Roget’s international thesaurus (6th ed.). New York, NY: Harper-Collins.

22.

Knopf

H. T.

Swick

K. J.

(2007). How parents feel about their child’s teacher/school: Implications for early childhood professionals. Early Childhood Education Journal, 34, 291-296.

23.

Kochanek

J. R.

(2005). Building trust for better schools: Research based practices. Thousand Oaks, CA: Corwin Press.

24.

Koufteros

Babbar

Kaighobadi

(2009). A paradigm for examining second-order factor models employing structural equation modeling. International Journal of Production Economics, 120, 633-652.

25.

Kramer

R. M.

(1999). Trust and distrust in organizations: Emerging perspectives, enduring questions. Annual Review of Psychology, 50, 569-598.

26.

Makiewicz

Mitchell

(2014). Teacher trust in the principal: Factor structure and effects. In VanMaele

Forsyth

P. B.

Van Houte

(Eds.), The role of trust for learning, teaching, leading, and bridging (pp. 99-120). Amsterdam, Netherlands: Springer.

27.

Marcoulides

G. A.

Heck

R. H.

(1993). Organizational culture and performance: Proposing and testing a model. Organization Science, 4, 209-225.

28.

Marsh

H. W.

Hau

Balla

J. R.

Grayson

(1998). Is more ever too much? The number of indicators per factor in confirmatory factor analysis. Multivariate Behavioral Research, 33, 181-220.

29.

Mayer

R. C.

Davis

J. H.

Schoorman

F. D.

(1995). An integrative model of organizational trust. Academy of Management Review, 20, 709-734.

30.

Mitchell

D. E.

(2015). Getting task structures and institutional designs right. In Michell

D. E.

Ream

R. K.

(Eds.), Professional responsibility: The fundamental issues in education and health care reform (pp. 175-194). Cham, Switzerland: Springer International Publishing.

31.

Mishra

A. K.

(1996). Organizational responses to crisis: The centrality of trust. In Kramer

Tyler

(Eds.), Trust in organizations: Frontiers of theory and research (pp. 261-287). Thousand Oaks, CA: Sage.

32.

Putnam

R. D.

(2001). Bowling alone: The collapse and revival of American community. New York, NY: Simon & Schuster.

33.

Raykov

(2009). Evaluation of scale reliability for unidimensional measures using latent variable modeling. Measurement and Evaluation in Counseling and Development, 42, 223-232.

34.

Rindskopf

Rose

(1988). Some theory and applications of confirmatory second-order factor analysis. Multivariate Behavioral Research, 23, 51-67.

35.

Romero

(2010). Student trust: Impacting high school outcomes (Unpublished doctoral dissertation). University of California Riverside.

36.

Romero

L. S.

(2015). Trust, behavior, and high school outcomes. Journal of Educational Administration, 53, 215-236.

37.

Schneider

Judy

Ebmeye

C. M.

Broda

(2014). Trust in elementary and secondary urban schools: A pathway for student success and college ambition. In VanMaele

Forsyth

Van Houtte

(Eds.), Trust and school life: The role of trust for learning, teaching, leading, and bridging (pp. 37-56). Dordrecht, Netherlands: Springer.

38.

Schoorman

F. D.

Mayer

R. C.

Davis

J. H.

(2007). An integrated model of organizational trust: Past, present, future. Academy of Management Review, 32, 344-354.

39.

Seashore-Louis

(2007). Trust and improvement in schools. Journal of Educational Change, 8, 1-24.

40.

Tschannen-Moran

(2004). Trust matters: Leadership for successful schools. San Francisco, CA: Jossey-Bass.

41.

Tschannen-Moran

(2009). Fostering teacher professionalism in schools: The role of leadership orientation and trust. Educational Administration Quarterly, 45, 217-247.

42.

Tschannen-Moran

Bankole

R. A.

Mitchell

R. M.

Moore

D. M.

Jr. (2013). Student academic optimism: A confirmatory factor analysis. Journal of Educational Administration, 51, 150-175.

43.

Tschannen-Moran

Gareis

C. R.

(2015). Faculty trust in the principal: An essential ingredient in high-performing schools. Journal of Educational Administration, 53, 66-92.

44.

VanMaele

Forsyth

P. B.

Van Houtte

(2014). Trust and school life. Dordrecht, Netherlands: Springer.