Psychometric Scale Development in Marketing Research

Scale to Measure Celebrity Endorsers

In a study by Ohanian, a scale used to measure celebrity endorsers' expertise, trustworthiness, and attractiveness is developed. Psychometric scale development protocols are followed for testing data reliability and validity. This study uses two exploratory and confirmatory samples to produce a 15-item scale measuring the characteristics of celebrity endorsers. The article complies fully with Churchill's recommendations on several fronts, as outlined below.

Several sources were researched to identify words, phrases, and adjectives for the questionnaire, resulting in the development of numerous adjectives describing personality traits. During the construction of the scale, 182 adjectives were identified, of which some were eliminated, reducing the list to approximately 139 adjectives. These 139 descriptors were further trimmed by a group of 38 college students — the researcher believed some words would be unfamiliar to respondents — down to 104.

For the identification of celebrities, 40 college students named celebrities they could recall within three minutes. The most frequently mentioned names were John McEnroe and Linda Evans, both of whom had been involved in advertisements. Celebrities frequently mentioned who had not been involved in advertisements included Tom Selleck and Madonna. Afterwards, a group of 38 students indicated their level of familiarity with each celebrity mentioned and named the most and least appropriate products those celebrities could sponsor. The results from these samples were used as guides for data collection, focusing on frequently purchased products used by most individuals. The final list established was: Linda Evans promoting a new perfume; Madonna, a new line of designer jeans; John McEnroe, tennis rackets; and Tom Selleck, a new line of men's cologne.

Factor Analysis and Scale Validation

The validation and purification of the data occurred in two phases: exploratory and confirmatory. All questionnaire items were factor analyzed using principal components analysis followed by varimax rotation. The initial factor solution for the sample evaluating Madonna resulted in four factors with eigenvalues greater than one. This four-factor solution accounted for 68.4% of the variance. To purify the list, items with loadings of 0.3 or greater on more than one factor were eliminated and the reduced list was factor analyzed again. The result was three factors with eigenvalues greater than one, accounting for 61% of the variance.

The final items from each factor analysis were tested for reliability using item-to-total correlations. The items for each subscale were analyzed separately. To obtain a practical scale size, items with the lowest item-to-total correlations were deleted while maintaining an acceptable level of reliability as measured by Cronbach's alpha (Ohanian, 1990). To determine the reliability of subscales across different celebrities and genders, Cronbach's alpha was computed for both male and female respondents for Madonna and John McEnroe. The results indicated a highly reliable scale. Both male and female respondents had equally reliable response patterns, and the total sample for each subscale and celebrity yielded a reliability coefficient of 0.8 or higher.

The factor loadings for selected scale items are presented below.

Factor Loadings — Madonna Sample (n = 249) and John McEnroe Sample (n = 237)

Authoritative: Madonna factors 1/2/3 = 0.617 / −0.370 / 0.051; McEnroe factors 1/2/3 = 0.654 / −0.156 / 0.009
Compatible with the Product: 0.795 / 0.103 / 0.070; 0.706 / −0.147 / 0.022
Expert: 0.765 / 0.107 / 0.045; 0.701 / −0.037 / 0.056
Informative: 0.812 / 0.170 / 0.183; 0.747 / 0.108 / 0.027
Experienced: 0.725 / 0.186 / 0.021; 0.531 / 0.077 / 0.034
Intelligent: 0.697 / 0.262 / 0.119; 0.699 / 0.133 / 0.054
Informed about the Product: 0.799 / 0.212 / 0.055; 0.631 / −0.025 / −0.195
Knowledgeable: 0.798 / 0.246 / 0.174; 0.695 / 0.121 / 0.079
Qualified: 0.869 / 0.069 / 0.131; 0.536 / 0.116 / −0.141

Using the LISREL methodology to verify the relationship between observable variables and latent constructs, two confirmatory factor analysis models — one for Linda Evans and one for Tom Selleck — were tested separately. The chi-square statistic was non-significant for each model, indicating an adequate fit of the confirmatory model to the data (χ²_Selleck = 99.60, df = 87, p = 0.168; χ²_Evans = 109.71, df = 87, p = 0.051). The root mean square residual was 0.048 and 0.046 for Selleck and Evans, respectively.

The scale demonstrates that it discriminates within and between respondents, yielding fair results from the data collected and clear analysis as recommended by Churchill (1979). Like any research, this study faces challenges. Being a quantitative analysis, it establishes the reliability and validity of the scale rather than discovering their existence. As such, the existing scale can undergo further modification. Given the large amounts of money spent on celebrity advertising, advertisers should use this scale as an integral part of their effectiveness testing and tracking.

The RELQUAL Scale for Export Relationship Quality

In a second publication, a RELQUAL scale is developed to assess the degree of relationship quality between exporting firms and importers. This article is also, to a considerable degree, compliant with the specifications issued by Churchill. The analysis explains whether reliability, data convergence, and validity are attainable from the instrument used for data collection. This study uses four multi-item scales and involved a sample of 1,564 British enterprises, of which 111 replies were returned in 2002, representing a response rate of 7%.

Churchill (1979) notes that for high reliability and low measurement error, multi-item scales rather than single-item scales should be used. This research uses Confirmatory Factor Analysis (CFA) with each item restricted to load on its pre-specified factor and the four first-order factors allowed to correlate freely. The chi-square for the collected data is significant (χ² = 125.97, df = 71, p < .05). Moreover, the Comparative Fit Index (CFI), Incremental Fit Index (IFI), and Tucker-Lewis Fit Index (TLI) for this model are 0.92, 0.92, and 0.90, respectively, indicating that the final structural model adequately reproduces the population covariance structure with an acceptable level of discrepancy between covariance matrices.

The items used in this scale are valid. All construct inter-correlations are significantly different from 1, while the square of their inter-correlations is less than the average variance extracted (Lages, Lages, & Lages, 2005). Discriminant validity is tested by including in the model an established construct (α = .83; vc(η) = .50; ρ = .83). Discriminant validity is further supported by non-significant correlations among the four first-order constructs and the new construct.

An EXPERF scale is also developed for use at the export venture level, with all three dimensions shown to be reliable and valid: financial export performance (α = .80; vc(η) = .61; ρ = .82), strategic export performance (α = .90; vc(η) = .77; ρ = .91), and satisfaction with the export venture (α = .90; vc(η) = .75; ρ = .90).

RELQUAL Correlations with Export Performance Dimensions

The table below presents path coefficients and t-values (in parentheses) for the relationships between RELQUAL constructs and export performance outcomes. * p < .05; ** p < .01 (two-tailed test).

Amount of information sharing in the relationship: Financial Export Performance = 0.39** (3.81); Strategic Export Performance = 0.30** (2.86); Satisfaction with Export Venture = 0.25* (2.40)
Communication quality in the relationship: 0.43** (4.53); 0.23* (2.22); 0.43** (4.82)
Long-term relationship orientation: 0.35** (3.48); 0.24* (2.34); 0.42** (4.60)
Satisfaction with the relationship: 0.71** (10.74); 0.39** (4.10); 0.73** (12.38)

This research encounters some challenges. The primary concern is that the final instrument may have caused common method variance, inflating the construct relationships. If common method bias exists, a CFA containing all constructs should produce a single method factor. However, the goodness-of-fit indices (CFI = .62, IFI = .63, TLI = .55) indicate a poor fit for the single-factor model, suggesting that biasing from common method variance is unlikely. A second limitation is the small sample size, which means the results should be considered suggestive rather than conclusive.

Nevertheless, when used to assess the qualities of a relationship, this scale helps managers better understand the dynamics of their business relationships. By defining strategies and actions that address potential problems with relationship quality, managers can ultimately influence their firm's performance.

Conclusion and Practical Implications

Across all three studies reviewed, adherence to Churchill's (1979) recommendations for developing better measures of marketing constructs produces scales with demonstrable reliability and construct validity. The Ohanian celebrity endorser scale, the RELQUAL export relationship quality scale, and the Zheng insurer trust scale each follow rigorous psychometric protocols — including multi-item scale construction, factor analysis, and Cronbach's alpha reliability testing — while acknowledging their respective methodological limitations.

Practically, these scales offer real value to decision-makers. Advertisers can use the celebrity endorser scale as an integral part of effectiveness testing. Export managers can apply the RELQUAL scale to diagnose and strengthen importer relationships. Health policy administrators and insurers can use the patient trust scale to identify where trust is eroding and to design interventions accordingly. Together, these studies illustrate how rigorous scale development, grounded in Churchill's paradigm, yields tools with both academic and applied utility.

References

Churchill, G. A. (1979). A paradigm for developing better measures of marketing constructs. Journal of Marketing Research, 16, 64–74.

Lages, C., Lages, C. R., & Lages, L. F. (2005). The RELQUAL scale: A measure of relationship quality in export market ventures. Lisbon, Portugal.

Ohanian, R. (1990). Construction and validation of a scale to measure celebrity endorsers' perceived expertise, trustworthiness, and attractiveness. Journal of Advertising, 19(3), 39–52.

Zheng, B., Hall, M. A., Kidd, K. E., & Levine, D. (2002). Development of a scale to measure patients' trust in health insurers. Health Services Research, 37(1), 187–202.