I have a question regarding data screening for an exploratory factor analysis (EFA).

I am conducting an EFA to identify the factor structure of 20 questions that I created on the topic of spirituality. I want to identify outliers in my sample using mahalanobis distances, and I am doing this on SPSS using a linear regression (Analyze -> Regression -> Linear).

  • I entered the 20 questions in SPSS as the "Independents", but what would variable should be entered under the "Dependent" category?
  • Or, if I can't use SPSS to find the mahalanobis distances, is there another (easy!!) way to find the m distances?
  • 2
    $\begingroup$ SPSS can compute Mahalanobis distances as a by-product in Linear regression and Discriminant analysis procedures. More convenient for you could be to use a special function to compute them. Take it from my web-page (Matrix - End Matrix functions). There are 2 functions for Mah. d. You'll need the second one, I guess. $\endgroup$ – ttnphns Aug 20 '12 at 7:02
  • 1
    $\begingroup$ P.S. @Madeline, in responce to your 1st question: in SPSS linear regression, specify any variable as as dependent (for example, respondent's ID number); check to save Mah.d. under Save button. That would be the most easy way for you. It will save that same squared distances as my function !smahalc computes. $\endgroup$ – ttnphns Aug 20 '12 at 8:26
  • $\begingroup$ Thank you ttnphns for you help! Also, I had to reverse code a few of the questions - does it matter if I enter the reverse coded or the "regular" responses when calculating mah. distances? $\endgroup$ – Madeline Aug 20 '12 at 10:03
  • $\begingroup$ I can't get exactly what you mean under "reverse code" but, anyway, you could do both ways to see if the result will change. $\endgroup$ – ttnphns Aug 20 '12 at 10:42
  • 1
    $\begingroup$ @ttnphns, Please post that as an answer. I would have thought you could simply use the PROXIMITIES command, but I see it is not an option. Good to know, I've never seen it used as a diagnostic tool for linear regression in my work so I would have never looked there. $\endgroup$ – Andy W Aug 20 '12 at 12:40

Here's a procedure:

  1. Create a dummy variable and move this variable to dependent variable box.
  2. Run linear regression process as usual.
  3. Save Mahalanobis score. (A new variable will be created in your data file.)

Extraordinarily large scores will be outliers.

  • 1
    $\begingroup$ Welcome to the site, @Muhammad-Rahid. I took the liberty of editing your answer to make it more readable & in-line w/ the sites policies, I hope you don't mind. Since you're new here, you may want to read our FAQ. $\endgroup$ – gung - Reinstate Monica Sep 28 '12 at 16:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.