Trust is one indicator of the quality of relationships with institutions and assesses dimensions not assessed by objective quality or subjective satisfaction measures. Trust is “the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the truster, irrespective of the ability to monitor or control that other party” (Mayer, Davis, and Schoorman 1995) or, more succinctly, “accepted vulnerability to another's possible but not expected ill will (or lack of good will)” (Baier 1986). Theorists have posited several forms of trust, two of which are especially pertinent to this discussion: interpersonal trust, which characterizes a relationship between two individuals, such as a specific doctor–patient relationship, and institutional trust, which characterizes a relationship with a collective, such as a corporation (Goold 1998; Hardin 1998; Luhmann 1988; Mechanic 1998a, 1998b).

There is a growing body of empirical work measuring the dimensions, levels, determinants, and consequences of trust in institutions such as business organizations (Kramer 1999; Kramer and Tyler 1996), governmental entities (Braithwaite and Levi 1998), financial institutions, and other commercial or social entities (Barber 1983; Govier 1997). Comparatively, the study of trust in the medical setting is still in its infancy. Research on interpersonal physician trust has shown that trust in primary care physicians is correlated with, but distinct from, satisfaction and improved patient outcomes (Anderson and Dedrick 1990; Kao, Green, Zaslavsky, et al. 1998; Safran, Kosinski, Tarlov, et al. 1998; Thom et al. 1999). The few studies assessing institutional trust in medicine (Blendon et al. 1998; Kao, Green, Zaslavsky, et al. 1998; Lake 2000) have relied on single-item measures of trust. This is in sharp contrast to the more robust multi-item scales used in trust studies with other institutions (Mishra and Spreitzer 1998).

This article reports on the development of a scale to measure trust in health insurers, including managed care organizations. The scale was developed and validated on a national random sample and was further validated and used in a regional sample of members in an HMO. Two principles guided the work reported here: First, we sought to develop a scale with a high level of validity, and second, we aimed for parsimony, believing a shorter scale would be more useful in research and practice.

Our model outlines the association between insurer trust and other constructs. First, insurer trust is related to physician trust, especially in managed care organizations (Mechanic and Schlesinger 1996). Patients who trust their physicians may worry less about their insurer because they count on their physicians to make appropriate referrals, to monitor the quality of services, or to provide the care that they need despite barriers imposed by the insurer. Similarly, trust in one's insurer may carry over to the health care professionals who are affiliated with that organization. Second, insurer trust is related to satisfaction. Theoretically, we distinguish trust from satisfaction by conceiving of trust as a psychological state of the subject that is primarily future oriented (“willingness to be vulnerable”) rather than an evaluation of past performance. Naturally, the two are associated. Finally, our conceptual model suggests insurer trust would be related to potential determinants or outcomes of trust, such as the type of insurer, how the insurer was selected, prior disputes with the insurer, and a desire to switch insurers.

These items were field tested and revised through four rounds of piloting, with a total sample of 290 male and female adults from various community groups (e.g., jury pool, health fair participants, airport passengers, university students, and clinic patients), representing a range of socioeconomic backgrounds. During this process, items were modified or deleted if there was a high rate of “don't knows” or if the answers were concentrated in one or two adjacent response categories, indicating a lack of discriminatory power. Data from 83 of these subjects, collected during the final two rounds of piloting, were analyzed to determine preliminary factor structure, internal consistency, and item-to-scale correlations. Items were rejected if they were weakly correlated with the overall scale or the relevant subscale.

Based on these iterations of content review, field testing, modification, and statistical analysis, 19 candidate items were selected for use in a national telephone survey, and a subset of 14 that performed well in the national sample were later used in a telephone survey of a regional sample of HMO members. Table 1 lists the 19 items. They cover the four dimensions of insurer trust: fidelity (items 3 and 15), competence (items 4, 9, and 12), confidentiality (items 8, 10, and 19), and honesty (items 1, 6, 7, and 17), as well as “global trust” items (items 2, 11, 13, 14, 16, and 18). For greater rigor in construct validity, most items do not use the word “trust” or its cognates such as “confidence.” We attempted to construct an instrument that measures trust based on an independent conceptualization rather than one that depends heavily on the subjects' internal definition.

Table 1 Nineteen Items and Their Mean, Standard Deviation, and Item-to-Total Correlation

Eighty-eight individuals did not meet the first criterion, and 131 individuals did not meet the second. Contacts with the 1,144 potentially eligible individuals resulted in the following dispositions: 564 (49.3 percent) were interviewed; 305 (26.7 percent) refused, and 275 (24 percent) were unable to complete the survey (not home, ill, not English speaking). A minimum of 15 attempts was made to those numbers that consistently were not answered. The fielding period for this study was April through June 1999. Telephone interviews lasted approximately 25 minutes. Data were collected on trust in the subject's regular physician or medical care provider, trust in the subject's current health insurer, demographic characteristics, satisfaction with care, and physical and mental health.

Data used in the following analyses were from 410 (73 percent) adults who answered a subset of 18 of the 19 insurer trust questions. (One question was deleted at the outset because of a high rate of nonresponse.) These subjects were younger than those who missed 1 or more of the 18 questions (mean age=47.6 vs. 61.2) but did not differ significantly with respect to gender, race, income, mental or physical health, the amount of contact with the health insurer, length of coverage with the insurer, satisfaction, patient trust, whether they belonged to a managed care plan, and the amount of choice they had in selecting their insurer.

Other variables thought to be related to insurer trust were measured as follows: the amount of choice one had in selecting an insurer (any: yes, no; enough: yes, no); past disagreement or dispute with the insurer (yes, no); desire to switch insurers (SA to SDA); and whether a subject belongs to a managed care plan. In the national sample, an insurer is considered managed care if at least two of these three possible attributes are described: prior authorization, provider network, or gatekeeping (Blendon et al. 1998).

All of the previously mentioned measures were collected in the national sample, whereas the regional sample included the insurer trust questions, enough choice in picking the insurer, prior dispute with insurer, and desire to switch insurer.

Table 2 Demographic Characteristics of National and Regional Samples (n = 410 and 1152)

These disparities do not exist in the regional HMO sample, in which 45.5 percent of subjects were male, and 14.1 percent were non-White. In other respects, the regional sample characteristics are substantially the same as in the national sample (Table 2).

In analyzing the national sample data, question 15 was deleted at the outset because of its high rate of nonresponse (n = 50). The remaining items had a much lower nonresponse rate (n < 20 for majority, n = 20 for items 7 and 12, and n = 24 for item 8), and acceptable response patterns, with standard deviations ranging from .75 to 1.14, and item means ranging from 2.61 to 3.81.

In the factor analysis, squared multiple correlations (between 0.15 and 0.69) were used as the prior commonality estimates. We found one factor with an eigenvalue (7.3) of greater than 1 that explained approximately 92 percent of the variance. The rest of the eigenvalues were less than 0.6. Therefore, we retained the single factor solution. Cronbach's alpha based on the 18 items was 0.92.

To reduce the length of the scale and preserve reliability, we deleted four items (8, 12, 13, and 19 in Table 1) with substantially lower item- to-total correlations than the others. The remaining 14 items were used in both the national and regional samples. Factor analysis found one factor in each sample, with eigenvalues 6.6 and 5.8, that explained approximately 96 percent and 97 percent of the variability in each sample, respectively. Cronbach's alpha was 0.92 and 0.91 in the two samples.

In our quest for parsimony, we deleted three additional items (2, 5, and 14) that were redundant or had a slightly more skewed frequency distribution. The remaining 11 items represent all of the domains of insurer trust (honesty=1, 7, 17; competence=4, 9; fidelity=3; confidentiality=10; and global=11, 14, 16, 18). This scale and each of the multi-item domains have a fairly even balance between positively (1, 4, 14, 17, 18) and negatively (3, 7, 9, 10, 11, 16) worded items.

Factor analysis of the 11 items still found one factor in both samples with eigenvalue 5.9 (national) and 4.8 (regional), which explained approximately 98 percent (national) and 99 percent (regional) of the variability. Cronbach's alpha in each sample was 0.92 and 0.89. Deletion of additional items would result in a Cronbach's alpha lower than our goal of 0.9 or would omit important content or both; thus, we retained the 11-item scale.

Insurer trust is measured by the sum of the 11 item scores (reverse-scored for negative items), ranging from 11 to 55, with a higher score indicating more trust. In the national sample, insurer trust had a mean of 36.5, a standard deviation of 7.8, and a range of 11 to 55. In the regional sample, insurer trust had a mean of 37.0, a standard deviation of 7.0, and a range of 12 to 55. The 2-month test–retest reliability in the regional sample was 0.76.

Table 3 Correlations in National Sample Among Insurer Trust, Physician Trust, Satisfaction, and Intent to Switch Insurer

In the national sample, physician trust is associated with both general (r = 0.52, p = 0.0001) and insurer satisfaction (s=0.19, p = 0.0001). General satisfaction with health care is correlated with insurer satisfaction (s=0.40, p = 0.0001) and desire to switch insurers (s=0.40, p = 0.0001). Finally, as expected, insurer satisfaction is correlated with intent to switch insurers (s=0.64, p = 0.0001).

Table 4 displays the sample size and the group mean (± SD) for insurer trust, physician trust, and general satisfaction in the national sample for the binary variables, and the Spearman correlations between insurer satisfaction and these variables. In both samples, having enough choice in selecting one's insurer (p < 0.01) and absence of dispute with the insurer (p = 0.0001) are associated with a higher level of insurer trust. In the national sample, they are also associated with a higher level of general satisfaction and insurer satisfaction. In the national sample, membership in a managed care plan is associated with lower levels of insurer trust (p = 0.0001), general satisfaction (p = 0.0001), and insurer satisfaction (s=0.19, p = 0.0007).

Table 4 Sample Size and Mean ± Standard Deviation of Insurer Trust, Physician Trust, General Satisfaction, and Spearman Correlation with Insurer Satisfaction for Binary Variables in National (and Regional) Samples

Trust is vital to relationships with individuals and institutions and may mediate critically important outcomes. Patients' relationships with and vulnerability to health insurers have assumed tremendous significance in the medical policy arena. Health insurers seek to instill public trust and thus avoid the heavy burdens of regulation that result from a loss of trust. Our findings indicate that trust is a strong predictor of member satisfaction and desire to remain with an insurer. In other institutional contexts, higher trust has been shown to increase willingness to reduce consumption in situations of limited resources (Brann and Foddy 1987; Kramer and Tyler 1996), which is especially relevant to managed care. Therefore, an instrument to measure trust in health insurers has utility as both a research tool and a management tool to monitor and improve performance.

The multi-item scale presented here applies to all types of health insurance and has adequate psychometric properties when used in a general population. The 11-item instrument has a consistent single-factor structure, high reliability and good construct validity, and moderate scale and item means. These statistical properties were observed in two distinct samples, one composed of a mix of private, public, indemnity, and managed care insurance and the other composed entirely of private HMO subscribers. However, this instrument has not yet been tested in large samples of more specialized populations, such as Medicaid or Medicare enrollees, or with non-English speakers.

Factor analyses indicate insurer trust is a unidimensional construct, in contrast to our starting assumption, but is consistent with findings from instruments measuring trust in physicians (Kao, Green, Zaslavsky, et al. 1998; Safran, Kosinski, Tarlov, et al. 1998; Thom et al. 1999). This means people do not appear to distinguish trust among dimensions of fidelity, confidentiality, competence, and honesty. This unidimensional conceptual model is further confirmed by the fact that the most global item (item 18, “all in all, you have complete trust in [insert insurer name]”) has by far the highest correlation to the overall scale (.81 and .79) and it is the only item that uses the term trust. This provides confirmation that the construct measured is trust. It is notable that although insurer trust is unidimensional, each of the dimensions we postulated remains in the final list of items.

It is also notable, however, that items measuring confidentiality (8, 10, 19) had among the lowest item-to-scale correlations (.42, .56, .43) and had the highest means (3.50, 3.48, 3.81) among the negatively worded items (which tend to produce lower means because of acquiescence bias) (Ware 1978). This suggests confidentiality is not a serious concern for most adults and the relationship between confidentiality and insurer trust is not as strong as expected. This is consistent with findings from research on physician trust (Kao, Green, Zaslavsky, et al. 1998; Thom et al. 1999).

Insurer trust exhibits a strong association with insurer satisfaction, with a desire to switch insurers, with having enough choice in selecting insurance, and with no prior disagreement with the insurer. Also, insurer trust appears substantially lower among managed care subscribers. These correlations are all consistent with prevailing conceptual theories of trust. It is notable the overlap between trust and satisfaction is much stronger for the single-item measure of satisfaction, which directly assesses satisfaction in the insurer, than with the less direct measure of general satisfaction with care.

A limitation of this study is that non-English speakers were excluded; thus, the study results and the insurer trust scale apply to English-speaking populations only. Another limitation is the validation measures are all self-reported attitudes, events, and predicted behaviors, rather than objective measures. Additional research is needed to determine the causative factors of trust and whether differences and changes in trust affect behaviors and other outcomes of interest.