Babeş-Bolyai University
Discussion
Started 8th Jan, 2020
Acceptable explained variance in exploratory factor analysis?
In exploratory factor analysis, is there a minimum variance that is considered to be acceptable?
Most recent answer
I'm not sure if this is what you are looking for, but:
(The reported) Eigenvalue = the total variance accounted to each factor. If for factor selection you use K1 (Kaiser criterion) or screeplot, minimum eigenvalue should be 1, or around 1 for scree. (However, a much better factor selection procedure is Parallel Analysis).
In addition, keep in mind that Eigenvalue alone is not enough to have a good representation on factor relevance; Watch for the cumulative amount of variance explained by the factors and for the uniqueness of the variables Vs factors, as well. (The greater the uniqueness, less relevant the variable is in the factor).
Popular replies (1)
Mississippi State University (Emeritus)
Hello Somipam,
The answer depends on your research goal. There are lots of guidelines from which to choose, but those are, in the end, just guidelines.
If your goal is instrument development, then you'd generally like the chosen factor structure to account for as much variance as possible (by discarding, revising, or replacing variables that do not have salient affiliations with specific factor/s). Very low values (insert your chosen threshold here!) suggest you haven't done a very good job in operationalizing the indicators for your target latent variable(s).
If it is for identifying factors/subscales that will have maximum correlation with external variables (criterion-related validity), then you might actually find that a specific cut-off for variance accounted for in the EFA could work against your efforts.
If the goal is to maximize estimates of internal consistency reliability of scores, then: (a) unidimensional structure is preferable to multidimensional structure; and (b) higher variable-factor loadings are best. Sticking to these principles will tend to maximize the resultant internal consistency estimates (though, you're now essentially cherry-picking, so another data set would be called for in order to estimate score reliability).
If the goal is data condensation (replacing a lot of variables with a smaller number of factors or components; or possibly to avoid concerns about collinearity), then you'd generally like to balance a higher variance accounted for relative to the requisite number of dimensions needed to meet that goal. For example, if 32 variables require 22 dimensions in order to explain 70% of the variance, then perhaps the combining of variables wasn't such a good idea...
Good luck with your work.
3 Recommendations
All replies (4)
Mississippi State University (Emeritus)
Hello Somipam,
The answer depends on your research goal. There are lots of guidelines from which to choose, but those are, in the end, just guidelines.
If your goal is instrument development, then you'd generally like the chosen factor structure to account for as much variance as possible (by discarding, revising, or replacing variables that do not have salient affiliations with specific factor/s). Very low values (insert your chosen threshold here!) suggest you haven't done a very good job in operationalizing the indicators for your target latent variable(s).
If it is for identifying factors/subscales that will have maximum correlation with external variables (criterion-related validity), then you might actually find that a specific cut-off for variance accounted for in the EFA could work against your efforts.
If the goal is to maximize estimates of internal consistency reliability of scores, then: (a) unidimensional structure is preferable to multidimensional structure; and (b) higher variable-factor loadings are best. Sticking to these principles will tend to maximize the resultant internal consistency estimates (though, you're now essentially cherry-picking, so another data set would be called for in order to estimate score reliability).
If the goal is data condensation (replacing a lot of variables with a smaller number of factors or components; or possibly to avoid concerns about collinearity), then you'd generally like to balance a higher variance accounted for relative to the requisite number of dimensions needed to meet that goal. For example, if 32 variables require 22 dimensions in order to explain 70% of the variance, then perhaps the combining of variables wasn't such a good idea...
Good luck with your work.
3 Recommendations
Babeş-Bolyai University
I'm not sure if this is what you are looking for, but:
(The reported) Eigenvalue = the total variance accounted to each factor. If for factor selection you use K1 (Kaiser criterion) or screeplot, minimum eigenvalue should be 1, or around 1 for scree. (However, a much better factor selection procedure is Parallel Analysis).
In addition, keep in mind that Eigenvalue alone is not enough to have a good representation on factor relevance; Watch for the cumulative amount of variance explained by the factors and for the uniqueness of the variables Vs factors, as well. (The greater the uniqueness, less relevant the variable is in the factor).
Similar questions and discussions
What is the communality cut-off value in EFA?
While performing EFA using Principal Axis Factoring with Promax rotation, Osborne, Costello, & Kellow (2008) suggests the communalities above 0.4 is acceptable. This is also suggested by James Gaskin on http://statwiki.kolobkreations.com . However, Child. D (2006) suggests that the value of communlaity below 0.2 should be removed.
In my case, the communalities are as low as 0.3 but inter-item correlation is above 0.3 as suggested by Field. A, (2009). What is the cut-off point for keeping an item based on the communality? Any other literature supporting (Child. D, 2006)?
Related Publications
The aim of the study is to perform an Exploratory Factor Analysis and to find the factors, which affect the success of the Producer Company. The study explains the factors. which are most influential for the success of the producer company. The study was carried out in Lathur Block of Kancheepuram district, Tamil Nadu in India, and consists of 200...
This paper aims to explain the main elements influencing the perceived value of the flamenco performer. In the framework of cultural economics, it presents a methodology based on two stages: interviews to experts, to identify the different aspects which influence the value of the performer; and surveys to consumers, to measure the valuation of stat...
The paper reports an investigation of whether sums of squared factor loadings obtained in confirmatory factor analysis correspond to eigenvalues of exploratory factor analysis. The sum of squared factor loadings reflects the variance of the corresponding latent variable if the variance parameter of the confirmatory factor model is set equal to one....