Login  Register

Re: about to exploratory factor analysis

Posted by Hector Maletta on Jan 28, 2012; 1:26am
URL: http://spssx-discussion.165.s1.nabble.com/about-to-exploratory-factor-analysis-tp5435719p5437191.html

The factors you get depend on the variables you use. Consequently, the distribution of total variance of all variables among the various factors depends on what variables you use. Each variables is regarded as standardized with variance=1, and this implies that k variables have total variance=k. If you drop one variable, total variance drops to k-1. This total variance, whatever it is, is distributed among the various factors, of which you can obtain up to the same number of observable variables (k or k-1 in the above examples). Once you drop one or more variables, factor analysis starts all over again, from scratch, taking only into account the remaining variables. Remember that factors are not objectively existing entities, but mathematical constructs that are linear functions of observed variables.

 

Hector

 

De: Arif OZER [mailto:[hidden email]]
Enviado el: Friday, January 27, 2012 17:26
Para: Hector Maletta; [hidden email]
Asunto: RE: about to exploratory factor analysis

 

Hi all,

Thanks for replies, but I think that I didnt express myself in the previous e-mails. Actually my question is that whether it is necessary to repeat exploratory factor analysis to determine the total variance, after the first analysis that is dropped variables or we must use to report the proportion of total variance obtained from  the first analysis?

thanks a lot again.

Yrd. Doç. Dr. Arif OZER
Gazi Universitesi
Meslek Eğitim Fakültesi
Eğitim Bilimleri Bölümü
Rehberlik ve Psikolojik Danışmanlık Anabilim Dalı
Beşevler / Ankara
Cep: 0506 287 72 65
iş:
e-posta: [hidden email]; [hidden email]; [hidden email]

 

> From: [hidden email]


> To: [hidden email]; [hidden email]
> Subject: RE: about to exploratory factor analysis
> Date: Fri, 27 Jan 2012 14:09:38 -0300
>
> The explained variance proportion is an attribute of the underlying
> components or factors. Loadings on each factor is an attribute of the
> observed variables, and can be changed by rotation of the factors.
> Discarding variables with lower loadings is not an analytical requirement
> but a heuristic decision, dictated by the desire to have a more frugal model
> with the minimum possible number of variables. If some variable appears to
> have low loadings, especially in the first few factors, and this is not
> greatly changed via rotation, eliminating those variables may be a smart
> move.
> A general warning, however, is that even variables that have low loadings in
> the first factor extracted may have a heavier weight in some other factor.
> Sometimes the problem is not defined by several observed variables mostly
> explained by the first factors (as in the classical case of several
> inter-correlated cognitive tests explained by the first factor or "general
> intelligence"). Sometimes, indeed, observed variables are better explained
> by several of the underlying factors, and it may also be the case that the
> applicable theory justifies this multi-factor analysis. Thus a variable with
> low loading on Factor 1 may have a larger loading on Factor 2 or 3.
> Besides, in some cases several variables, having load loadings individually,
> may make, all together, a substantial contribution. If these variables (or
> some of them) somehow can be interpreted as belonging in the same underlying
> dimension (e.g. by being all correlated to the same factor, even if that
> factor is not the first) or if they reflect variables with important
> theoretical functions, then retaining them may be wise. For instance,
> suppose in an instance of the same classical analysis of cognitive ability
> you have some variable (or variables) reflecting the nervousness of the
> subjects in the test situation: even if that variable (or variables) have a
> relatively low loading on the first factor, they may be important to retain.
>
> Finally, it may be the case that through rotation of factors you may get a
> better (and more easily interpretable) picture of the importance of those
> variables.
>
> Hector
>
> -----Mensaje original-----
> De: SPSSX(r) Discussion [[hidden email]] En nombre de
> [hidden email]
> Enviado el: Friday, January 27, 2012 09:33
> Para: [hidden email]
> Asunto: about to exploratory factor analysis
>
> Hi all,
>
> I have a question. Namely, running exploratory factor analysis (EFA) in
> spss, it gives explained variance proportion for each eiganvalue. Besides,
> the variables that have loading less than .30 can be dropped out from data
> set according to this analysis. Herein, reporting the total explained
> variance in articles, is it necessary or appropriate to repeat the EFA with
> reduced data set and report the total explained variance obtained from
> second analysis?
> thanks in advance
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command SIGNOFF SPSSX-L For a list of
> commands to manage subscriptions, send the command INFO REFCARD
> -----
> No virus found in this message.
> Checked by AVG - www.avg.com
> Version: 2012.0.1901 / Virus Database: 2109/4768 - Release Date: 01/26/12
>


No virus found in this message.

Checked by AVG - www.avg.com
Version: 2012.0.1901 / Virus Database: 2109/4770 - Release Date: 01/27/12