+ Reply to Thread
Results 1 to 4 of 4

Thread: Exploratory factor analysis with forced number of factors?

  1. #1
    Points: 746, Level: 14
    Level completed: 46%, Points required for next Level: 54

    Posts
    12
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Exploratory factor analysis with forced number of factors?




    Dear colleagues,

    SPSS has an option for exploratory factor analysis[/SIZE] where you can choose the number of factors before performing the analysis. In fact, you can force a specific number of factors on the data. I donít understand this. If it is exploratory then why can we force the analysis to have a specified number of factors? How should one use this option?
    Thanks for your help.

    Anthony

  2. #2
    Instagram oscar.olvera100
    Points: 18,877, Level: 87
    Level completed: 6%, Points required for next Level: 473
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,079
    Thanks
    164
    Thanked 520 Times in 414 Posts

    Re: Exploratory factor analysis with forced number of factors?

    Quote Originally Posted by Anthony_1000 View Post
    Dear colleagues,

    SPSS has an option for exploratory factor analysis[/SIZE] where you can choose the number of factors before performing the analysis. In fact, you can force a specific number of factors on the data. I don’t understand this. If it is exploratory then why can we force the analysis to have a specified number of factors? How should one use this option?
    Thanks for your help.

    Anthony
    Exploratory Factor Analysis gets the 'exploratory' out of the fact that you do not specify in advance which factor loads on which items. All factors are allowed to load on all items. You always, always need to specify the number of factors because, if you do not, then your model is not identified and you cannot obtain parameter estimates.

    Whenever you don't ask SPSS to select the number of factors, SPSS uses a flawed, problematic rule that was popular in the last century where only eigenvalues greater than 1 are selected. But whether you formally choose the number of factors or you let SPSS do it, the number of factors always has to be specified in advance.

    In my opinion, one as a researcher should always select the number of factors. There is a wide variety of 'pointers' to probe the number of factors (Parallel Analysis, Velicer's Minimum Average Partial correlations, etc.) which you can use to inform your judgement, but nothing will ever trump the theory that substantiates the purpose behind your analysis. Latent variable models in psychometrics are mostly theory-driven and it is important not to rely on heuristics or quick-n-easy solutions to fundamental problems of research.
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  3. #3
    Points: 746, Level: 14
    Level completed: 46%, Points required for next Level: 54

    Posts
    12
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Exploratory factor analysis with forced number of factors?

    Thank you Spunky for your reply,

    Iím not very much familiar with this way of doing FA. Are there fit statistics to evaluate the number of factors we propose? If a researcher is doubtful about the number of factors and runs the FA with 1, 2, 3, etc. number of factors in several analyses then how can they choose the best model?

    Best
    Anthony

  4. #4
    Instagram oscar.olvera100
    Points: 18,877, Level: 87
    Level completed: 6%, Points required for next Level: 473
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,079
    Thanks
    164
    Thanked 520 Times in 414 Posts

    Re: Exploratory factor analysis with forced number of factors?


    Quote Originally Posted by Anthony_1000 View Post
    Thank you Spunky for your reply,

    I’m not very much familiar with this way of doing FA. Are there fit statistics to evaluate the number of factors we propose? If a researcher is doubtful about the number of factors and runs the FA with 1, 2, 3, etc. number of factors in several analyses then how can they choose the best model?

    Best
    Anthony
    Yes, there are. The most-often recommended "strategy" (not statistic/hypothesis test) to decide on the number of factors is a simulation-based approach called Parallel Analysis. More modern psychometric theory (and software), however, allows you to conduct series of chi-square tests of fit with models of differing numbers of factors to see which one fits the data better. It would also be possible (depending on the estimator) to look at information criteria of various types (AIC, BIC, DIC) at different sample sizes to see which one minimizes misfit.

    But the most important thing to point out here is that no statistical trickery can trump solid theory in the process of building a scale/questionnaire. By the time you've reached the stage of doing Factor Analysis you already need to be fairly certain of the number of dimensions that are being measured by your scale. If you aren't, you need to go back and start from scratch because you're probably not measuring the construct(s) that you're interested in measuring properly anyway.
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats