Please any one can tell me the basic difference between these technique and why we use maximum likelihood with promax incase of EFA before conducting confirmatory factor analysis by AMOS?
With principal component analysis and rotation: You can do an oblique rotation first (oblimin, promax), and examine the component correlation matrix. If the correlation between the components is not important, you can repeat the analysis with varimax rotation. The latter will maximize the factor loadings, assuming no correlations between components. If there are important correlations between the components, then you can save the factor scores and perform a second-order principal component analysis on them. OR, you can discuss the component correlations themselves, depending on the research question.
Maximum Likelihood Estimation is a technique employed in confirmatory factor analysis to estimate parameters for a model, in which you specify explicitly the expected relations between the factors and the endogenous variables. Then you examine the goodness-of-fit of your specifications. Instead of rotating, you may delete paths with small coefficients and repeat the process, until the fit improves.
I agree with Rimantas....you can only use rotation if you are doing EFA and not CFA.
If your factors are not correlated, employ varimax rotation, other wise promax or other techniques, especially if your factors are significantly correlated.
Varimax rotation is orthogonal rotation in which assumption is that there is no intercorrelations between components. Promax rotation requires large data set usually < 150. If you hav small data set, you can use oblimin rotation. You can use maximum likelihood extraction in EFA but the data should be normally distributed.
I'm currently doing EFA and not sure if I should be using direct oblimin or promax rotation - would anyone be able to advise, please? My sample size is 127, but others in my field doing EFA on the same scale have used promax. However, they have larger sample sizes of 300+.
Please guide.
When Oblique rotation are appropriate for a factor model in which a common factor are assumed to be (independent or dependent) ?
Where an orthogonal simple structure rotation is desired, varimax should be applied. Maximum likelihood (ML) analysis differs from principal components in a number of ways. Successive factoring explains as much variance as possible in a population correlation matrix. Thus, ML is suited for confirmatory analysis. However, when test reliabilities, and thus communalities are high, the difference between ML and principal components is trivial. This means a more simple solution like components analysis may be desirable. I have not discussed when to use different rotation methods since this did not appear to be your major question. I am responding to your first question. The previous comments are
excellent.
First, in oblique rotations, the factor axes can take up any position in factor space. As you know, the cosine of the angle between the factor axes indicates the correlation between them. Since the axes can take up any position in the real world, you need to know your data well enough to choose the correct value.
Also, as delta controls the degree of correlation among the rotated factors, delta values of zero yield factors that are highly correlated. Large negative values produce factors that are more uncorrelated. Too many may not know their data well enough and just take the default value in SPSS. The optimum value will depend on how well you know your data and the study.
You may want to check the delta range? If one overrides zero, then
the range is -0.8 to + 0.8. The optimum kappa value is that value
that gives the simplest structure with the lowest correlations among
factors. The default value of 4 was recommended by its developers
(Hendrickson and White) as generally providing the optimum solution
(raising the power of the loadings).
I found this article really helpful, please check: https://hosted.jalt.org/test/PDF/Brown31.pdf
Hope able help some!
- Badges
- Science topic
- Similar topics
- Filing