Sei sulla pagina 1di 2

Confirmatory Factor Analysis (CFA)

Many a time I have seen academics and students doing an Exploratory Factor Analysis (EFA)
and then follow up with a Confirmatory Factor Analysis (CFA) with the same data set.
Although I have been stressing many times many people are still not convinced or do not buy
the idea.
Here I am putting down the evidence from the book by Hair et al. (2010), the excerpt is
shown below. Here they have argued that after the 5th Stage ie; Label the Factors, Stage 6 is
the validation stage. Here they have clearly stated that to validate the measures you can
follow 2 strategies:
1. Split the sample with the original data, the first data set to develop the measures,
and the second data set to validate; or
2. Collect a separate new sample to do the validation.
This has also been clearly demonstrated in their earlier book, what they did back in Hair et al.
(2006) was to run an EFA with a sample of 100 pilot test sample and then validate the
purified sample with a new set of sample which was 400. Never have they suggested running
the EFA with the data set and the running a CFA to confirm. As it has been clearly stated
that in EFA, we do not know the factor structure (let the computer decide) whereas in CFA
we already know and we want to confirm.

References

Hair J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate Data Analysis.
Upper Saddle River, NJ: Prentice-Hall.
Hair J. F., Black, W. C., Babin, B. J., & Anderson, R. E. & Tatham R. L. (2006). Multivariate
Data Analysis. Upper Saddle River, NJ: Prentice-Hall.
According to Kline (2005), page 204 he states, it is not entirely appropriate to specify a CFA
model based on the results of the exploratory factor analysis (EFA) and to estimate the former
using the same data. That is, the CFA would not in this case confirm the results of the EFA.
He further adds, this is because EFA results are subject to capitalization on chance variation,
and specification of a CFA model based on the EFA outcomes and analyzed with the same
data may just compound this problem. Besides, there is some evidence that factor structures
identified in EFA may turn out to have a poor fit to the same data when evaluated with CFA.
This may be especially likely when all relatively high factor loadings from EFA are specified
as free parameters in CFA, but all relatively low (secondary) EFA factor loadings (e.g., those
with absolute values less than 0.39) are fixed to zero in CFA (van Prooijen & van der Kloot,
2001). Secondary loadings in EFA often account for relatively high proportions of the
variance, so constraining them to zero in CFA may be too conservative. It is preferable to
validate a factor structure across a different sample and to use the same method, either EFA or
CFA, in both samples.
Kline, R. B. (2005). Principles and Practice of Structural Equation Modeling (2nd ed.). New
York: Guilford.
Van Prooijen, J-W. & van der Kloot, W. A. (2001). Confirmatory Analysis of Exploratively
Obtained Factor Structures. Educational and Psychological Measurement, 61(5), 777792.

Potrebbero piacerti anche