Confirmatory Analysis of Exploratively Obtained Factor Structures

Abstract

Factor structures obtained by exploratory factor analysis (EFA) often turn out to fit poorly in confirmative follow-up studies. In the present study, the authors assessed the extent to which results obtained in EFA studies can be replicated by confirmatory factor analysis (CFA) in the same sample. More specifically, the authors used CFA to test three different factor models on several correlation matrices of exploratively obtained factor structures that were reported in the literature. The factor models varied with respect to the role of the smaller factor pattern coefficients. Results showed that confirmatory factor models in which all low EFA pattern coefficients were fixed to zero fitted especially poorly. The authors conclude that it may be justified to use a less constrained model when testing a factor model by allowing some correlation among the factors and some of the lower factor pattern coefficients to differ from zero.Multifactor psychological scales and tests are very common measurement tools that are widely used in research and in practical applications. Psychological knowledge is, for a large part, founded on such psychometric instruments. Most multifactor tests and measurement instruments are initially developed by exploratory factor analysis (EFA) (Gorsuch, 1983), which produces a set of factor pattern coefficients on, usually, orthogonal factors based on the correlations between the items of the test. If one wants to investigate whether the factor structure of the test can be replicated in a new study, one could, in principle, perform EFA on the new data. Subsequently, one

The Turkish Version of Web-Based Learning Platform Evaluation Scale: Reliability and Validity Study

“…CFA can also be applied in studies of scale translation from a different language in order to verify the predetermined factor structure (Çokluk et al, 2012, p. 283). Owing to the impossibility of a comparison of EFA and CFA, on the basis of the inappropriate EFA applications and cultural differences (Van Prooijen & Van der Kloot, 2001), the factor structure formed by EFA may not be verified with the data received from a different sample (e.g. Dunn et al, 2006;Hooper, Marotta, & Depuy, 2009;Öztürk, 2011).…”

“…Dunn et al, 2006;Hooper, Marotta, & Depuy, 2009;Öztürk, 2011). In order to examine such cases, Van Prooijen and Van der Kloot (2001) stated that conducting CFA of the factor structures on the same dataset can be more effective.…”

“…Thus, although the validity and reliability analyses of the original scale were performed with fewer samples than those in the present study, EFA was still applied to examine the factor structure of the Turkish version of the scale. Moreover, in order to support the structure formed by EFA, confirmatory factor analysis (CFA) was conducted on the same sample (Kline, 1994;Van Prooijen & Van der Kloot, 2001). …”

Confirmatory Factor Analysis of Employee Trust Scale

“…Confirmatory Factor Analysis (CFA) was chosen for the present study as it can help researchers to test a model that can suggest in advance the relationship between latent factors and observed variables along with the relationship among the factors as well. This type of model is sensitive to the parameters such as being fixed to a certain value within a model, remaining equal to other parameters, and being independent while taking in any unknown value [10].…”

Evaluating the implementation and sustainability of an evidence-based intervention: delivering Triple P within Indigenous child welfare agencies

“…e Both FIML and multiple imputations are considered the most efficient methods for handling missing data, when the data is at least MAR. When conducting both a CFA and EFA it is not appropriate to use the same sample of participants for the two different factor analyses (Van Prooijen & Van der Kloot, 2001). Therefore, the data file was randomly split into two samples to conduct EFA (n = 292) and CFA (n = 300).…”