Implementing alternative estimation methods to test the construct validity of Likert-scale instruments.

A manuscript recently published in Nursing Research [1] suggested using polychoric correlations and polychoric confirmatory factor analysis (CFA) for unbiased assessments of construct validity in Likert-scale instruments, rather than Pearson correlations and Pearson correlation-based CFA. An editorial in the most recent issue of Psychological Test Adoption and Development also recommended the weighted least square mean and variance-adjusted (WLSMV) method for CFA-based validity testing [2]. Using polychoric correlation for CFA involves applying CFA estimation methods to ordinal item variables. However, relatively few nursing studies have used this estimation method to test the construct validity of ordinal variables. As a general recommendation, the maximum likelihood (ML) method can be used for instruments with 5 to 7 item categories, as seen in the Likert scales commonly employed in nursing research [3]. The frequent application of strict cutoff rules for model fit indices to evaluate construct validity based on CFA estimation results may lead to an underestimation of the study instrument and modification of the CFA model by removing items or introducing connected item residual terms. Therefore, better assessment methods of the construct validity of Likert scales are needed, and alternative estimation methods are recommended to avoid incorrect parameter estimates, such as factor loading coefficients, standard errors, and model fit statistics [4]. In this context, the purpose of this paper is to explain the necessity of alternative estimation methods and to present how those methods can be applied using affordable, accessible, and appropriate structural equation modeling (SEM) programs.